We present CrystalDiT, a diffusion transformer for crystal structure generation that achieves state-of-the-art performance by challenging the trend of architectural complexity. Instead of intricate, multi-stream designs, CrystalDiT employs a unified transformer that imposes a powerful inductive bias: treating lattice and atomic properties as a single, interdependent system. Combined with a periodic table-based atomic representation and a balanced training strategy, our approach achieves 8.78% SUN (Stable, Unique, Novel) rate on MP-20, substantially outperforming recent methods including FlowMM (4.21%) and MatterGen (3.66%). Notably, CrystalDiT generates 63.28% unique and novel structures while maintaining comparable stability rates, demonstrating that architectural simplicity can be more effective than complexity for materials discovery. Our results suggest that in data-limited scientific domains, carefully designed simple architectures outperform sophisticated alternatives that are prone to overfitting.

Large language model (LLM) agents have shown remarkable progress in social deduction games (SDGs).However, existing approaches primarily focus on information processing and strategy selection, overlooking the significance of persuasive communication in influencing other players' beliefs and responses.In SDGs, success depends not only on making correct deductions but also on convincing others to respond in alignment with one's intent.To address this limitation, we formalize turn-based dialogue in SDGs as a Stackelberg competition, where the current player acts as the leader who strategically influences the follower's response.Building on this theoretical foundation, we propose a reinforcement learning framework that trains agents to optimize utterances for persuasive impact.Through comprehensive experiments across four diverse social deduction benchmarks, we demonstrate that our agents significantly outperform baselines.This work represents a significant step toward developing AI agents capable of strategic social influence, with implications extending to scenarios requiring persuasive communication.

MOTIVATION: Traditional drug discovery methods are costly and inefficient, while existing deep learning approaches remain limited by task specificity and practical applicability. Accurately modeling protein-molecule interactions is critical for advancing virtual screening, docking, and drug design. RESULTS: We propose DrugBLIP, a multi-task graph transformer model based on SE(3)-equivariant architectures, to unify protein-molecule interaction learning. By integrating contrastive learning, matching tasks, and docking optimization, DrugBLIP captures 3D spatial relationships through a hybrid graph transformer framework. Evaluations demonstrate state-of-the-art performance: DrugBLIP achieves an AUROC of 0.8217 and BEDROC of 0.5743 on virtual screening, outperforming traditional and deep learning baselines by 10%-127% across metrics. It also attains 91.2% top-1 docking success on CASF-2016 and 41.8% target fishing accuracy, showcasing robustness in diverse scenarios. Additionally, DrugBLIP reduces computational time by 700× compared to traditional docking tools. AVAILABILITY AND IMPLEMENTATION: Code is available at https://github.com/Wolkenwandler/DrugBLIP and archived at Zenodo with DOI: 10.5281/zenodo.16990700.

Discrete Token Modeling (DTM), which employs vector quantization techniques, has demonstrated remarkable success in modeling non-natural language modalities, particularly in time series generation. While our prior work SDformer established the first DTM-based framework to achieve state-of-the-art performance in this domain, two critical limitations persist in existing DTM approaches: 1) their inability to capture multi-scale temporal patterns inherent to complex time series data, and 2) the absence of theoretical foundations to guide model optimization. To address these challenges, we proposes a novel multi-scale DTM-based time series generation method, called Multi-Scale Discrete Transformer (MSDformer). MSDformer employs a multi-scale time series tokenizer to learn discrete token representations at multiple scales, which jointly characterize the complex nature of time series data. Subsequently, MSDformer applies a multi-scale autoregressive token modeling technique to capture the multi-scale patterns of time series within the discrete latent space. Theoretically, we validate the effectiveness of the DTM method and the rationality of MSDformer through the rate-distortion theorem. Comprehensive experiments demonstrate that MSDformer significantly outperforms state-of-the-art methods. Both theoretical analysis and experimental results demonstrate that incorporating multi-scale information and modeling multi-scale patterns can substantially enhance the quality of generated time series in DTM-based approaches.

RGB-Thermal multi-object tracking (RGB-T MOT) focuses on tracking multiple objects in complex scenarios, such as nighttime and low-light conditions, which is crucial for various applications, including video surveillance and unmanned aerial vehicle monitoring. Existing RGB-T MOT studies typically merge multi-source features in a single stage before the backbone network. These works fail to preserve fine-grained details and struggle with context-dependent fusion, resulting in suboptimal tracking performance. In this paper, we propose a novel multi-stage cross-modality spatial-temporal feature interaction network (MCTrack), which emphasizes two main aspects: temporal-aware learning and cross-modality information interaction. For temporal-aware learning, we introduce the temporal salient feature interaction (TSFI) module, which ensures trajectory continuity by capturing dynamic spatial variation of objects across RGB-T video pairs, enabling consistent object tracking over time. For cross-modality information interaction, we propose the bidirectional modality interaction (Bi-MI) module, which employs cross-modality transformers to extract complementary features from both RGB and thermal modalities at multiple stages of feature extraction, thereby improving tracking adaptability in diverse tracking scenarios. Additionally, we propose the cross-modality complementary mask (CCM) strategy, which applies non-overlapping random masks to feature maps to improve the robustness of cross-modality feature interaction. Extensive experiments and comparative analyses demonstrate that our MCTrack surpasses state-of-the-art trackers on both publicly available VT-MOT and UniRTL-MOT RGB-T datasets. The code is available at https://github.com/ydhcg-BoBo/RGB-T-MOT.

Generative models have gained significant attention in multivariate time series forecasting (MTS), particularly due to their ability to generate high-fidelity samples. Forecasting the probability distribution of multivariate time series is a challenging yet practical task. Although some recent attempts have been made to handle this task, two major challenges persist: 1) some existing generative methods underperform in high-dimensional multivariate time series forecasting, which is hard to scale to higher dimensions; 2) The inherent high-dimensional multivariate attributes constrain the forecasting lengths of existing generative models. In this paper, we point out that discrete token representations can model high-dimensional MTS with faster inference time, and forecast the target with the long-term trends of itself can extend the forecasting length with high accuracy. Motivated by this, we propose a vector quantized framework called Hierarchical Discrete Transformer (HDT) that models time series into discrete token representations with l2 normalization enhanced vector quantized strategy, in which we transform the MTS forecasting into discrete tokens generation. To address the limitations of generative models in long-term forecasting, we propose a hierarchical discrete Transformer. This model captures the discrete long-term trend of the target at the low level and leverages this trend as a condition to generate the discrete representation of the target at the high level that introduces the features of target itself for extending the forecasting length in high-dimensional MTS. Extensive experiments on five popular MTS datasets verify the effectiveness of our proposed method. The source code will be released.

Vision-language foundation models have exhibited remarkable success across a multitude of downstream tasks due to their scalability on extensive image-text paired data. However, these models also display significant limitations when applied to downstream tasks, such as fine-grained image classification, as a result of ``decision shortcuts'' that hinder their generalization capabilities. In this work, we find that the CLIP model possesses a rich set of features, encompassing both desired invariant causal features and undesired decision shortcuts. Moreover, the underperformance of CLIP on downstream tasks originates from its inability to effectively utilize pre-trained features in accordance with specific task requirements. To address this challenge, we propose a simple yet effective method, Spurious Feature Eraser (SEraser), to alleviate the decision shortcuts by erasing the spurious features. Specifically, we introduce a test-time prompt tuning paradigm that optimizes a learnable prompt, thereby compelling the model to exploit invariant features while disregarding decision shortcuts during the inference phase. The proposed method effectively alleviates excessive dependence on potentially misleading spurious information. We conduct comparative analysis of the proposed method against various approaches which validates the significant superiority.

Spiking Neural Networks (SNNs) offer a promising, biologically inspired approach for processing spatiotemporal data, particularly for time series forecasting. However, conventional neuron models like the Leaky Integrate-and-Fire (LIF) struggle to capture long-term dependencies and effectively process multi-scale temporal dynamics. To overcome these limitations, we introduce the Temporal Segment Leaky Integrate-and-Fire (TS-LIF) model, featuring a novel dual-compartment architecture. The dendritic and somatic compartments specialize in capturing distinct frequency components, providing functional heterogeneity that enhances the neuron's ability to process both low- and high-frequency information. Furthermore, the newly introduced direct somatic current injection reduces information loss during intra-neuronal transmission, while dendritic spike generation improves multi-scale information extraction. We provide a theoretical stability analysis of the TS-LIF model and explain how each compartment contributes to distinct frequency response characteristics. Experimental results show that TS-LIF outperforms traditional SNNs in time series forecasting, demonstrating better accuracy and robustness, even with missing data. TS-LIF advances the application of SNNs in time-series forecasting, providing a biologically inspired approach that captures complex temporal dynamics and offers potential for practical implementation in diverse forecasting scenarios.

Diffusion Language Models (DLMs) have emerged as a promising new paradigm for text generative modeling, potentially addressing limitations of autoregressive (AR) models. However, current DLMs have been studied at a smaller scale compared to their AR counterparts and lack fair comparison on language modeling benchmarks. Additionally, training diffusion models from scratch at scale remains challenging. Given the prevalence of open-source AR language models, we propose adapting these models to build text diffusion models. We demonstrate connections between AR and diffusion modeling objectives and introduce a simple continual pre-training approach for training diffusion models. Through systematic evaluation on language modeling, reasoning, and commonsense benchmarks, we show that we can convert AR models ranging from 127M to 7B parameters (GPT2 and LLaMA) into diffusion models DiffuGPT and DiffuLLaMA, using less than 200B tokens for training. Our experimental results reveal that these models outperform earlier DLMs and are competitive with their AR counterparts. We release a suite of DLMs (127M-355M-7B) capable of generating fluent text, performing in-context learning, filling in the middle without prompt re-ordering, and following instructions.

Hallucination remains a significant challenge in Large Vision-Language Models (LVLMs). To alleviate this issue, some methods, known as contrastive decoding, induce hallucinations by manually disturbing the raw vision or instruction inputs and then mitigate them by contrasting the outputs of the original and disturbed LVLMs. However, these holistic input disturbances sometimes induce potential noise and also double the inference cost. To tackle these issues, we propose a simple yet effective method named $\textit{Self-Introspective Decoding}$ (SID). Our empirical investigations reveal that pre-trained LVLMs can introspectively assess the importance of vision tokens based on preceding vision and text (both instruction and generated) tokens. Leveraging this insight, we develop the Context and Text-aware Token Selection (CT$^2$S) strategy, which preserves only the least important vision tokens after the early decoder layers, thereby adaptively amplify vision-and-text association hallucinations during auto-regressive decoding. This strategy ensures that multimodal knowledge absorbed in the early decoder layers induces multimodal contextual rather than aimless hallucinations, and significantly reduces computation burdens. Subsequently, the original token logits subtract the amplified fine-grained hallucinations, effectively alleviating hallucinations without compromising the LVLMs' general ability. Extensive experiments illustrate SID generates less-hallucination and higher-quality texts across various metrics, without much additional computation cost.

Immunoglobulins are crucial proteins produced by the immune system to identify and bind to foreign substances, playing an essential role in shielding organisms from infections and diseases. Designing specific antibodies opens new pathways for disease treatment. With the rise of deep learning, AI-driven drug design has become possible, leading to several methods for antibody design. However, many of these approaches require additional conditions that differ from real-world scenarios, making it challenging to incorporate them into existing antibody design processes. Here, we introduce IgGM, a generative model for the de novo design of immunoglobulins with functional specificity. IgGM simultaneously generates antibody sequences and structures for a given antigen, consisting of three core components: a pre-trained language model for extracting sequence features, a feature learning module for identifying pertinent features, and a prediction module that outputs designed antibody sequences and the predicted complete antibody-antigen complex structure. IgGM effectively predicts structures and designs novel antibodies and nanobodies. This makes it highly applicable in a wide range of practical situations related to antibody and nanobody design. Code is available at: https://github.com/TencentAI4S/IgGM.

Machine learning models must continuously self-adjust themselves for novel data distribution in the open world. As the predominant principle, entropy minimization (EM) has been proven to be a simple yet effective cornerstone in existing test-time adaption (TTA) methods. While unfortunately its fatal limitation (i.e., overconfidence) tends to result in model collapse. For this issue, we propose to \textbf{\texttt{Co}}nservatively \textbf{\texttt{M}}inimize the \textbf{\texttt{E}}ntropy (\texttt{COME}), which is a simple drop-in replacement of traditional EM to elegantly address the limitation. In essence, \texttt{COME} explicitly models the uncertainty by characterizing a Dirichlet prior distribution over model predictions during TTA. By doing so, \texttt{COME} naturally regularizes the model to favor conservative confidence on unreliable samples. Theoretically, we provide a preliminary analysis to reveal the ability of \texttt{COME} in enhancing the optimization stability by introducing a data-adaptive lower bound on the entropy. Empirically, our method achieves state-of-the-art performance on commonly used benchmarks, showing significant improvements in terms of classification accuracy and uncertainty estimation under various settings including standard, life-long and open-world TTA, i.e., up to $34.5\%$ improvement on accuracy and $15.1\%$ on false positive rate. Our code is available at: \href{https://github.com/BlueWhaleLab/COME}{https://github.com/BlueWhaleLab/COME}.

Spiking Neural Networks (SNNs) often suffer from high time complexity $O(T)$ due to the sequential processing of $T$ spikes, making training computationally expensive. In this paper, we propose a novel Fixed-point Parallel Training (FPT) method to accelerate SNN training without modifying the network architecture or introducing additional assumptions. FPT reduces the time complexity to $O(K)$, where $K$ is a small constant (usually $K=3$), by using a fixed-point iteration form of Leaky Integrate-and-Fire (LIF) neurons for all $T$ timesteps. We provide a theoretical convergence analysis of FPT and demonstrate that existing parallel spiking neurons can be viewed as special cases of our approach. Experimental results show that FPT effectively simulates the dynamics of original LIF neurons, significantly reducing computational time without sacrificing accuracy. This makes FPT a scalable and efficient solution for real-world applications, particularly for long-duration simulations.

The alignment of large language models (LLMs) often assumes that using more clean data yields better outcomes, overlooking the match between model capacity and example difficulty. Challenging this, we propose a new principle: *Preference data vary in difficulty, and overly difficult examples hinder alignment, by exceeding the model's capacity*. Through systematic experimentation, we validate this principle with three key findings: (1) preference examples vary in difficulty, as evidenced by consistent learning orders across alignment runs; (2) overly difficult examples significantly degrade performance across four LLMs and two datasets; and (3) the capacity of a model dictates its threshold for handling difficult examples, underscoring a critical relationship between data selection and model capacity. Building on this principle, we introduce *Selective DPO*, which filters out overly difficult examples. This simple adjustment improves alignment performance by 9-16\% in win rates on the AlpacaEval 2 benchmark compared to the DPO baseline, surpassing a series of DPO variants with different algorithmic adjustments. These results together illuminate the importance of aligning data difficulty with model capacity, offering a transformative perspective for improving alignment strategies in LLMs. Code is available at https://github.com/glorgao/SelectiveDPO

In this paper, we seek to develop a versatile test-time adaptation (TTA) objective for a variety of tasks — classification and regression across image-, object-, and pixel-level predictions. We achieve this through a self-bootstrapping scheme that optimizes prediction consistency between the test image (as target) and its deteriorated view. The key challenge lies in devising effective augmentations/deteriorations that: i) preserve the image’s geometric information, e.g., object sizes and locations, which is crucial for TTA on object/pixel-level tasks, and ii) provide sufficient learning signals for TTA. To this end, we analyze how common distribution shifts affect the image’s information power across spatial frequencies in the Fourier domain, and reveal that low-frequency components carry high power and masking these components supplies more learning signals, while masking high-frequency components can not. In light of this, we randomly mask the low-frequency amplitude of an image in its Fourier domain for augmentation. Meanwhile, we also augment the image with noise injection to compensate for missing learning signals at high frequencies, by enhancing the information power there. Experiments show that, either independently or as a plug-and-play module, our method achieves superior results across classification, segmentation, and 3D monocular detection tasks with both transformer and CNN models.

Inspired by the impressive capabilities of GPT-4o, there is growing interest in enabling speech language models (SLMs) to engage in natural, fluid spoken interactions with humans. Recent advancements have led to the development of several SLMs that demonstrate promising results in this area. However, current approaches have yet to fully exploit dual-channel speech data, which inherently captures the structure and dynamics of human conversation. In this work, we systematically explore the use of dual-channel speech data in the context of modern large language models, and introduce a novel generative modeling paradigm—Next-Token-Pair Prediction (NTPP)—to enable speaker-independent dual-channel spoken dialogue learning using decoder-only architectures for the first time. We evaluate our approach on standard benchmarks, and empirical results show that our proposed method, NTPP, significantly improves the conversational abilities of SLMs in terms of turn-taking prediction, response coherence, and naturalness. Moreover, compared to existing methods, NTPP achieves substantially lower inference latency, highlighting its practical efficiency for real-time applications. Demo and code can be found at https://audio-3059.pages.dev.

Offline reinforcement learning is widely applied in multiple fields due to its advantages in efficiency and risk control. However, a major problem it faces is the distribution shift between offline datasets and online environments. This mismatch leads to out-of-distribution (OOD) state-action pairs that fall outside the scope of the training data. Therefore, existing conservative training policies may not provide reliable decisions when the test environment deviates greatly from the offline dataset. In this paper, we propose Test-time Adapted Reinforcement Learning (TARL) to address this problem. TARL constructs unsupervised test-time optimization objectives for discrete and continuous control tasks, using test data without depending on environmental rewards. In discrete control tasks, it minimizes the entropy of predicted action probabilities to decrease uncertainty and avoid OOD state-action pairs. For continuous control tasks, it represents and minimizes action uncertainty based on the normal distribution of policy network outputs. Moreover, to prevent model bias caused by overfitting and error accumulation during the test-time update process, TARL enforces a KL divergence constraint between the fine-tuned policy and the original policy. For efficiency, TARL only updates the layer normalization layer parameters during testing. Extensive experiments on popular Atari game benchmarks and the D4RL dataset demonstrate the superiority of our method. Our method achieved a significant improvement over CQL, with a 13.6% episode return relative increase on the hopper-expert-v2 task.

Large language models (LLMs) are widely adopted to generate synthetic datasets for various natural language processing (NLP) tasks, such as text classification and summarization. However, accurately measuring the diversity of these synthetic datasets—an aspect crucial for robust model performance—remains a significant challenge. In this paper, we introduce DCScore, a novel method for measuring synthetic dataset diversity from a classification perspective. Specifically, DCScore formulates diversity evaluation as a sample classification task, leveraging mutual relationships among samples. We further provide theoretical verification of the diversity-related axioms satisfied by DCScore, highlighting its role as a principled diversity evaluation method. Experimental results on synthetic datasets reveal that DCScore enjoys a stronger correlation with multiple diversity pseudo-truths of evaluated datasets, underscoring its effectiveness. Moreover, both empirical and theoretical evidence demonstrate that DCScore substantially reduces computational costs compared to existing methods. Code is available at: https://github.com/bluewhalelab/dcscore.

Multi-task learning (MTL) has emerged as a promising approach for deploying deep learning models in real-life applications. Recent studies have proposed optimization-based learning paradigms to establish task-shared representations in MTL. However, our paper empirically argues that these studies, specifically gradient-based ones, primarily emphasize the conflict issue while neglecting the potentially more significant impact of imbalance/dominance in MTL. In line with this perspective, we enhance the existing baseline method by injecting imbalance-sensitivity through the imposition of constraints on the projected norms. To demonstrate the effectiveness of our proposed IMbalance-sensitive Gradient (IMGrad) descent method, we evaluate it on multiple mainstream MTL benchmarks, encompassing supervised learning tasks as well as reinforcement learning. The experimental results consistently demonstrate competitive performance.

Implied Volatility Surface (IVS) prediction is critical for options hedging, portfolio management, and risk control. These applications encounter significant challenges, including imbalanced data distributions and inherent uncertainties in forecasting. Recent advances relying on deep learning have led to significant progress in addressing these issues. However, several key problems have not yet been solved: (i) Moneyness and maturities of traded options change over time. Therefore, proper spatio-temporal alignment is an essential prerequisite for the downstream forecasting exercise. (ii) Different regions of the IVS are unevenly informed because of liquidity constraints and therefore should not be modeled uniformly. (iii) The complex interconnections among data points in the IVS from various perspectives-such as the well-known 'smirk' patterns across dimensions-are neither explicitly addressed nor effectively captured by existing models. To address these issues, we propose a novel end-to-end heterogeneous cross(x)-view aligned graph attention network (Hexagon-Net), which aligns historical IVS data, learns distinctive IVS patterns, propagates predictive information, and forecasts future IVS movements simultaneously. Extensive experiments on stock index options datasets demonstrate that Hexagon-Net significantly and consistently outperforms the previous approaches in IVS modeling and deep learning. Additionally, we present further experiments-such as ablation studies, sensitivity analyses, and alternative configurations-to explore the reasons behind its superior performance.

Existing neural methods for multi-task vehicle routing problems (VRPs) typically learn unified solvers to handle multiple constraints simultaneously. However, they often underutilize the compositional structure of VRP variants, each derivable from a common set of basis VRP variants. This critical oversight causes unified solvers to miss out the potential benefits of basis solvers, each specialized for a basis VRP variant. To overcome this limitation, we propose a framework that enables unified solvers to perceive the shared-component nature across VRP variants by proactively reusing basis solvers, while mitigating the exponential growth of trained neural solvers. Specifically, we introduce a State-Decomposable MDP (SDMDP) that reformulates VRPs by expressing the state space as the Cartesian product of basis state spaces associated with basis VRP variants. More crucially, this formulation inherently yields the optimal basis policy for each basis VRP variant. Furthermore, a Latent Space-based SDMDP extension is developed by incorporating both the optimal basis policies and a learnable mixture function to enable the policy reuse in the latent space. Under mild assumptions, this extension provably recovers the optimal unified policy of SDMDP through the mixture function that computes the state embedding as a mapping from the basis state embeddings generated by optimal basis policies. For practical implementation, we introduce the Mixture-of-Specialized-Experts Solver (MoSES), which realizes basis policies through specialized Low-Rank Adaptation (LoRA) experts, and implements the mixture function via an adaptive gating mechanism. Extensive experiments conducted across VRP variants showcase the superiority of MoSES over prior methods.

With the development of biotechnology, RNA therapies have shown great potential. However, different from proteins, the sequences corresponding to a single RNA three-dimensional structure are more abundant. Most of the existing RNA design methods merely take into account the secondary structure of RNA, or are only capable of generating a limited number of candidate sequences. To address these limitations, we propose a geometric-algebra-enhanced $\textbf{B}$ayesian $\textbf{F}$low $\textbf{N}$etwork for the inverse design of $\textbf{R}$NA, called $\textbf{RBFN}$. RBFN uses a Bayesian Flow Network to model the distribution of nucleotide sequences in RNA, enabling the generation of more reasonable RNA sequences. Meanwhile, considering the more flexible characteristics of RNA conformations, we utilize geometric algebra to enhance the modeling ability of the RNA three-dimensional structure, facilitating a better understanding of RNA structural properties. In addition, due to the scarcity of RNA structures and the limitation that there are only four types of nucleic acids, we propose a new time-step distribution sampling to address the scarcity of RNA structure data and the relatively small number of nucleic acid types. Evaluation on the single-state fixed-backbone re-design benchmark and multi-state fixed-backbone benchmark indicates that RBFN can outperform existing RNA design methods in various RNA design tasks, enabling effective RNA sequence design.

Existing methods to enhance the reasoning capability of large language models predominantly rely on supervised fine-tuning (SFT) followed by reinforcement learning (RL) on reasoning-specific data. These approaches critically depend on external supervisions--such as labeled reasoning traces, verified golden answers, or pre-trained reward models. In this work, we propose Entropy Minimized Policy Optimization (EMPO), which makes an early attempt at fully unsupervised LLM reasoning incentivization. By minimizing the semantic entropy of LLMs on unlabeled questions, EMPO achieves competitive performance compared to supervised counterparts. Specifically, without any supervised signals, EMPO boosts the accuracy of Qwen2.5-Math-7B Base from 33.7\% to 51.6\% on math benchmarks and improves the accuracy of Qwen2.5-7B Base from 32.1\% to 50.1\% on MMLU-Pro. Primary analysis are also provided to interpret the effectiveness of EMPO. Code is available at https://github.com/QingyangZhang/EMPO.

Multi-task learning (MTL) is a widely explored paradigm that enables the simultaneous learning of multiple tasks using a single model. Despite numerous solutions, the key issues of optimization conflict and task imbalance remain under-addressed, limiting performance. Unlike existing optimization-based approaches that typically reweight task losses or gradients to mitigate conflicts or promote progress, we propose a novel approach based on Continual Optimization with Symmetry Teleportation (COST). During MTL optimization, when an optimization conflict arises, we seek an alternative loss-equivalent point on the loss landscape to reduce conflict. Specifically, we utilize a low-rank adapter (LoRA) to facilitate this practical teleportation by designing convergent, loss-invariant objectives. Additionally, we introduce a historical trajectory reuse strategy to continually leverage the benefits of advanced optimizers. Extensive experiments on multiple mainstream datasets demonstrate the effectiveness of our approach. COST is a plug-and-play solution that enhances a wide range of existing MTL methods. When integrated with state-of-the-art methods, COST achieves superior performance.

Nowadays deep models are required to be versatile due to the increasing realistic needs. Multi-task learning (MTL) offers an efficient way for this purpose to learn multiple tasks simultaneously with a single model. However, prior MTL solutions often focus on resolving conflicts and imbalances during optimization, which may not outperform simple linear scalarization strategies~\citep{xin2022current}. Instead of altering the optimization trajectory, this paper leverages mode connectivity to efficiently approach the Pareto front and identify the desired trade-off point. Unlike Pareto Front Learning (PFL), which aims to align with the entire Pareto front, we focus on effectively and efficiently exploring optimal trade-offs. However, three challenges persist: (1) the low-loss path can neither fully traverse trade-offs nor align with user preference due to its randomness, (2) commonly adopted Bézier curves in mode connectivity are ill-suited to navigating the complex loss landscapes of deep models, and (3) poor scalability to large-scale task scenarios. To address these challenges, we adopt non-uniform rational B-Splines (NURBS) to model mode connectivity, allowing for more flexible and precise curve optimization. Additionally, we introduce an order-aware objective to explore task loss trade-offs and employ a task grouping strategy to enhance scalability under massive task scenarios. Extensive experiments on key MTL datasets demonstrate that our proposed method, *EXTRA* (EXplore TRAde-offs), effectively identifies the desired point on the Pareto front and achieves state-of-the-art performance. *EXTRA* is also validated as a plug-and-play solution for mainstream MTL approaches.

One underlying assumption of recent Federated Learning (FL) paradigms is that all local models share an identical network architecture. However, this assumption is inefficient for heterogeneous systems where devices possess varying computation and communication capabilities. The presence of such heterogeneity among devices negatively impacts the scalability of FL and slows down the training process due to the existence of stragglers. To this end, this paper proposes a novel federated compression framework for heterogeneous models , named FedHM, distributing the heterogeneous low-rank models to clients and then aggregating them into a full-rank global model. Furthermore, FedHM significantly reduces communication costs by utilizing low-rank models. Compared with state-of-the-art heterogeneous FL methods under various FL settings, FedHM is superior in the performance and robustness of models with different sizes. Additionally, the convergence guarantee of FL for heterogeneous devices is first theoretically analyzed.

Test-time adaptation (TTA) seeks to tackle potential distribution shifts between training and testing data by adapting a given model w.r.t. any testing sample. This task is particularly important when the test environment changes frequently. Although some recent attempts have been made to handle this task, we still face two key challenges: 1) prior methods have to perform backpropagation for each test sample, resulting in unbearable optimization costs to many applications; 2) while existing TTA solutions can significantly improve the test performance on out-of-distribution data, they often suffer from severe performance degradation on in-distribution data after TTA (known as catastrophic forgetting). To this end, we have proposed an Efficient Anti-Forgetting Test-Time Adaptation (EATA) method which develops an active sample selection criterion to identify reliable and non-redundant samples for test-time entropy minimization. To alleviate forgetting, EATA introduces a Fisher regularizer estimated from test samples to constrain important model parameters from drastic changes. However, in EATA, the adopted entropy loss consistently assigns higher confidence to predictions even when the samples are underlying uncertain, leading to overconfident predictions that underestimate the data uncertainty. To tackle this, we further propose EATA with Calibration (EATA-C) to separately exploit the reducible model uncertainty and the inherent data uncertainty for calibrated TTA. Specifically, we compare the divergence between predictions from the full network and its sub-networks to measure the reducible model uncertainty, on which we propose a test-time uncertainty reduction strategy with divergence minimization loss to encourage consistent predictions instead of overconfident ones. To further re-calibrate predicting confidence on different samples, we utilize the disagreement among predicted labels as an indicator of the data uncertainty. Based on this, we devise a min-max entropy regularization to selectively increase and decrease predicting confidence for confidence re-calibration. Note that EATA-C and EATA are different on the adaptation objective, while EATA-C still benefits from the active sample selection criterion and anti-forgetting Fisher regularization proposed in EATA. Extensive experiments on image classification and semantic segmentation verify the effectiveness of our proposed methods.

Current video object segmentation methods heavily rely on pixel-level mask annotations when training, which are expensive and time-consuming to acquire. To address this problem, some approaches try to train with sparse scribble annotations and take sparse target scribble as initial information for inference. However, due to the sparsity of scribble annotations, the performance is often limited, and the corresponding loss function needs to be designed. Inspired by the powerful ability of Segment Anything Model (SAM) to leverage prompt for segmentation, we argue that this problem can be alleviated by improving the quality of scribble. Therefore, we propose SEVOS, a framework for scribble-supervised video object segmentation, which contains a scribble enhancement algorithm and an semi-supervised video object segmentation network. Specifically, the scribble enhancement algorithm first samples corresponding positive sample points and negative sample points from target scribbles, and then feeds them into the SAM in turn, achieving high-quality scribble enhancement without human intervention. This algorithm augments the scribble-annotated video dataset, which is used for additional training of the model. Furthermore, we design a post-processing enhancement algorithm to further improve the prediction results. The obtained model outperforms state-of-the-art methods with a considerable performance gap, indicating the generalization and effectiveness of the proposed model.

Learning a good similarity measure for large-scale high-dimensional data is a crucial task in machine learning applications, yet it poses a significant challenge. Distributed minibatch Stochastic Gradient Descent (SGD) serves as an efficient optimization method in large-scale distributed training, allowing linear speedup in proportion to the number of workers. However, communication efficiency in distributed SGD requires a sufficiently large minibatch size, presenting two distinct challenges. Firstly, a large minibatch size leads to high memory usage and computational complexity during parallel training of high-dimensional models. Second, a larger batch size of data reduces the convergence rate. To overcome these challenges, we propose an Efficient Distributed Sparse Relative Similarity Learning ( \(\mathbf{\mathsf{EDSRSL}}\) ) framework. This framework integrates two strategies: local minibatch SGD and sparse relative similarity learning. By effectively reducing the number of updates through synchronous delay while maintaining a large batch size, we address the issue of high computational cost. Additionally, we incorporate sparse model learning into the training process, significantly reducing computational cost. This article also provides theoretical proof that the convergence rate does not decrease significantly with increasing batch size. Various experiments on six high-dimensional real-world datasets demonstrate the efficacy and efficiency of the proposed algorithms, with a communication cost reduction of up to \(90.89\%\) and a maximum wall time speedup of \(5.66\times\) compared to the baseline methods.

Bandit online multiclass prediction plays an important role in many real-world applications. In this paper, we propose a unified Bandit Online Multiclass Prediction (BOMP) framework. This framework is based on our proposed margin-based gradient descent approach. Its update step provides an unbiased estimate of the surrogate loss gradient and has a lower variance than existing methods. It also enables our algorithms to update even for incorrect predictions by penalizing the wrong classes. The link function of the framework can evolve over time, gradually incorporating online data information including second-order information into the potential functions. Based on the proposed framework, we investigate first-order and second-order bandit online multiclass prediction algorithms. Theoretical analysis demonstrates the superiority of our proposed update rule and bandit online multiclass prediction framework. Finally, we compare our proposed first-order and second-order bandit online multiclass prediction algorithms with several state-of-the-art methods on two synthetic and four real-world datasets. The encouraging results show that our proposed algorithms significantly outperform state-of-the-art techniques.

The probability prediction of multivariate time series is a notoriously challenging but practical task. This research proposes to condense high-dimensional multivariate time series forecasting into a problem of latent space time series generation, to improve the expressiveness of each timestamp and make forecasting more manageable. To solve the problem that the existing work is hard to extend to high-dimensional multivariate time series, we present a latent multivariate time series diffusion framework called Latent Diffusion Transformer (LDT), which consists of a symmetric statistics-aware autoencoder and a diffusion-based conditional generator, to implement this idea. Through careful design, the time series autoencoder can compress multivariate timestamp patterns into a concise latent representation by considering dynamic statistics. Then, the diffusion-based conditional generator is able to efficiently generate realistic multivariate timestamp values on a continuous latent space under a novel self-conditioning guidance which is modeled in a non-autoregressive way. Extensive experiments demonstrate that our model achieves state-of-the-art performance on many popular high-dimensional multivariate time series datasets.

Liang Chen, Yatao Bian, Yang Deng, Deng Cai, Shuaiyi Li, Peilin Zhao, Kam-Fai Wong. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Missing values and unreleased figures are common but highly important for backtesting and real-time analysis in the financial industry, yet underexploited in the existing literature. In this paper, we focus on the issue of empirical asset pricing, where the cross-section of future asset returns is a function of lagged firm characteristics that vary in time frequencies and missing ratios. Most of the existing imputation methods cannot fully capture the complex and evolving spatio-temporal relations among firm-level characteristics. In particular, these methods fail to explicitly consider the spatial relations and feature structure in the stock network where we have to process granular data of thousands of stocks and hundreds of characteristics for each stock. To address these challenges, we propose a spatio-temporal diffusion model (STDM) that gradually recovers the masked financial data conditioning on high-dimensional stock-and-characteristics historical data. We propose characteristic-specific projection to construct characteristic-level features at both ends of the STDM, meanwhile maintaining firm-level features in the middle of the STDM to largely reduce the computational memory. Moreover, along with the temporal attention, we design a spatial graph convolutional network, making it computationally efficient and effective to learn time-varying spatio-temporal interdependence across firms. We further employ an implicit sampler that greatly accelerates the inference procedure so that the STDM is able to produce high-quality point and density estimates of missing and real-time firm characteristics within a few steps. We evaluate our model on the most comprehensive open-source dataset 'OSAP' and generate state-of-the-art performance in extensive experiments.

This paper explores the open research problem of understanding the social behaviors of LLM-based agents.Using Avalon as a testbed, we employ system prompts to guide LLM agents in gameplay.While previous studies have touched on gameplay with LLM agents, research on their social behaviors is lacking.We propose a novel framework, tailored for Avalon, features a multi-agent system facilitating efficient communication and interaction.We evaluate its performance based on game success and analyze LLM agents' social behaviors.Results affirm the framework's effectiveness in creating adaptive agents and suggest LLM-based agents' potential in navigating dynamic social interactions.By examining collaboration and confrontation behaviors, we offer insights into this field's research and applications.Our code is publicly available at https://github.com/ 3DAgentWorld/LLM-Game-Agent.

Graph Neural Networks (GNNs) with equivariant properties have emerged as powerful tools for modeling complex dynamics of multi-object physical systems. However, their generalization ability is limited by the inadequate consideration of physical inductive biases: (1) Existing studies overlook the continuity of transitions among system states, opting to employ several discrete transformation layers to learn the direct mapping between two adjacent states; (2) Most models only account for first-order velocity information, despite the fact that many physical systems are governed by second-order motion laws. To incorporate these inductive biases, we propose the Second-order Equivariant Graph Neural Ordinary Differential Equation (SEGNO). Specifically, we show how the second-order continuity can be incorporated into GNNs while maintaining the equivariant property. Furthermore, we offer theoretical insights into SEGNO, highlighting that it can learn a unique trajectory between adjacent states, which is crucial for model generalization. Additionally, we prove that the discrepancy between this learned trajectory of SEGNO and the true trajectory is bounded. Extensive experiments on complex dynamical systems including molecular dynamics and motion capture demonstrate that our model yields a significant improvement over the state-of-the-art baselines.

Learning neural subset selection tasks, such as compound selection in AI-aided drug discovery, have become increasingly pivotal across diverse applications. The existing methodologies in the field primarily concentrate on constructing models that capture the relationship between utility function values and subsets within their respective supersets. However, these approaches tend to overlook the valuable information contained within the superset when utilizing neural networks to model set functions. In this work, we address this oversight by adopting a probabilistic perspective. Our theoretical findings demonstrate that when the target value is conditioned on both the input set and subset, it is essential to incorporate an invariant sufficient statistic of the superset into the subset of interest for effective learning. This ensures that the output value remains invariant to permutations of the subset and its corresponding superset, enabling identification of the specific superset from which the subset originated. Motivated by these insights, we propose a simple yet effective information aggregation module designed to merge the representations of subsets and supersets from a permutation invariance perspective. Comprehensive empirical evaluations across diverse tasks and datasets validate the enhanced efficacy of our approach over conventional methods, underscoring the practicality and potency of our proposed strategies in real-world contexts.

Deep long-tailed recognition (DTLR) has attracted much attention due to its close touch with realistic scenarios. Recent advances have focused on re-balancing across various aspects, e.g., sampling strategy, loss re-weighting, logit adjustment, and input/parameter perturbation, to name a few. However, few studies have considered dynamic re-balancing to address intrinsic optimization conflicts. In this paper, we first empirically argue that the optimizations of mainstream DLTR methods are still dominated by some categories (e.g., major) due to a fixed re-balancing strategy. Thus, they fail to deal with gradient conflicts among categories, which naturally deduces the motivation for reaching Pareto optimal solutions. Unfortunately, a naive integration of multi-objective optimization (MOO) with DLTR methods is not applicable due to the gap between multi-task learning (MTL) and DLTR, and can in turn lead to class-specific feature degradation. Thus, we provide effective alternatives by decoupling MOO-based MTL from the temporal rather than structure perspective, and enhancing it via optimizing variability collapse loss motivated by the derived MOO-based DLTR generalization bound. Moreover, we resort to anticipating worst-case optimization with theoretical insights to further ensure convergence. We build a Pareto deep long-tailed recognition method termed PLOT upon the proposed MOO framework. Extensive evaluations demonstrate that our method not only generally improves mainstream pipelines, but also achieves an augmented version to realize state-of-the-art performance across multiple benchmarks.

Test-time adaptation has proven effective in adapting a given trained model to unseen test samples with potential distribution shifts. However, in real-world scenarios, models are usually deployed on resource-limited devices, e.g., FPGAs, and are often quantized and hard-coded with non-modifiable parameters for acceleration. In light of this, existing methods are often infeasible since they heavily depend on computation-intensive backpropagation for model updating that may be not supported. To address this, we propose a test-time Forward-Optimization Adaptation (FOA) method. In FOA, we seek to solely learn a newly added prompt (as model’s input) via a derivative-free covariance matrix adaptation evolution strategy. To make this strategy work stably under our online unsupervised setting, we devise a novel fitness function by measuring test-training statistic discrepancy and model prediction entropy. Moreover, we design an activation shifting scheme that directly tunes the model activations for shifted test samples, making them align with the source training domain, thereby further enhancing adaptation performance. Without using any backpropagation and altering model weights, FOA runs on quantized 8-bit ViT outperforms gradient-based TENT on full-precision 32-bit ViT, while achieving an up to 24-fold memory reduction on ImageNet-C. The source code is available at: https://github.com/mr-eggplant/FOA.

Catastrophic forgetting remains a core challenge in continual learning (CL), where the models struggle to retain previous knowledge when learning new tasks. While existing replay-based CL methods have been proposed to tackle this challenge by utilizing a memory buffer to store data from previous tasks, they generally overlook the interdependence between previously learned tasks and fail to encapsulate the optimally integrated knowledge in previous tasks, leading to sub-optimal performance of the previous tasks. Against this issue, we first reformulate replay-based CL methods as a unified hierarchical gradient aggregation framework. We then incorporate the Pareto optimization to capture the interrelationship among previously learned tasks and design a Pareto-Optimized CL algorithm (POCL), which effectively enhances the overall performance of past tasks while ensuring the performance of the current task. Comprehensive empirical results demonstrate that the proposed POCL outperforms current state-of-the-art CL methods across multiple datasets and different settings.

Industrial multimedia recommendation systems extensively utilize cascade architectures to deliver personalized content for users, generally consisting of multiple stages like retrieval and ranking. However, retrieval models have long suffered from Sample Selection Bias (SSB) due to the distribution discrepancy between the exposed items used for model training and the candidates (almost unexposed) during inference, affecting recommendation performance. Traditional methods utilize retrieval candidates as augmented training data, indiscriminately treating unexposed data as negative samples, which leads to inaccuracies and noise. Some efforts rely on unbiased datasets, while they are costly to collect and insufficient for industrial models. In this paper, we propose a debiasing framework named DAMCAR, which introduces Domain Adaptation to mitigate SSB in Multimedia CAscade Recommendation systems. Firstly, we sample hard-to-distinguish samples from unexposed data to serve as the target domain, optimizing data quality and resource utilization. Secondly, adversarial domain adaptation is employed to generate pseudo-labels for each sample. To enhance robustness, we utilize Exponential Moving Average (EMA) to create a teacher model that supervises the generation of pseudo-labels via self-distillation. Finally, we obtain a retrieval model that maintains stable performance during inference through a hybrid training mechanism. We conduct offline experiments on two real-world datasets and deploy our approach in the retrieval model of a multimedia video recommendation system for online A/B testing. Comprehensive experimental results demonstrate the effectiveness of DAMCAR in practical applications.

The superior generation capabilities of Denoised Diffusion Probabilistic Models (DDPMs) have been effectively showcased across a multitude of domains. Recently, the application of DDPMs has extended to time series generation tasks, where they have significantly outperformed other deep generative models, often by a substantial margin. However, we have discovered two main challenges with these methods: 1) the inference time is excessively long; 2) there is potential for improvement in the quality of the generated time series. In this paper, we propose a method based on discrete token modeling technique called Similarity-driven Discrete Transformer (SDformer). Specifically, SDformer utilizes a similarity-driven vector quantization method for learning high-quality discrete token representations of time series, followed by a discrete Transformer for data distribution modeling at the token level. Comprehensive experiments show that our method significantly outperforms competing approaches in terms of the generated time series quality while also ensuring a short inference time. Furthermore, without requiring retraining, SDformer can be directly applied to predictive tasks and still achieve commendable results.

Recently, Large Language Models (LLMs) have garnered significant attention in recommendation systems, improving recommendation performance through in-context learning or parameter-efficient fine-tuning. However, cross-domain generalization, i.e., model training in one scenario (source domain) but inference in another (target domain), is underexplored. In this paper, we present TLRec, a transfer learning framework aimed at enhancing LLMs for sequential recommendation tasks. TLRec specifically focuses on text inputs to mitigate the challenge of limited transferability across diverse domains, offering promising advantages over traditional recommendation models that heavily depend on unique identities (IDs) like user IDs and item IDs. Moreover, we leverage the source domain data to further enhance LLMs’ performance in the target domain. Initially, we employ powerful closed-source LLMs (e.g., GPT-4) and chain-of-thought techniques to construct instruction tuning data from the third-party scenario (source domain). Subsequently, we apply curriculum learning to fine-tune LLMs for effective knowledge injection and perform recommendations in the target domain. Experimental results demonstrate that TLRec achieves superior performance under the zero-shot and few-shot settings.

MOTIVATION: Retrosynthesis is a critical task in drug discovery, aimed at finding a viable pathway for synthesizing a given target molecule. Many existing approaches frame this task as a graph-generating problem. Specifically, these methods first identify the reaction center, and break a targeted molecule accordingly to generate the synthons. Reactants are generated by either adding atoms sequentially to synthon graphs or by directly adding appropriate leaving groups. However, both of these strategies have limitations. Adding atoms results in a long prediction sequence that increases the complexity of generation, while adding leaving groups only considers those in the training set, which leads to poor generalization. RESULTS: In this paper, we propose a novel end-to-end graph generation model for retrosynthesis prediction, which sequentially identifies the reaction center, generates the synthons, and adds motifs to the synthons to generate reactants. Given that chemically meaningful motifs fall between the size of atoms and leaving groups, our model achieves lower prediction complexity than adding atoms and demonstrates superior performance than adding leaving groups. We evaluate our proposed model on a benchmark dataset and show that it significantly outperforms previous state-of-the-art models. Furthermore, we conduct ablation studies to investigate the contribution of each component of our proposed model to the overall performance on benchmark datasets. Experiment results demonstrate the effectiveness of our model in predicting retrosynthesis pathways and suggest its potential as a valuable tool in drug discovery. AVAILABILITY AND IMPLEMENTATION: All code and data are available at https://github.com/szu-ljh2020/MARS.

Binding affinity prediction of three-dimensional (3D) protein-ligand complexes is critical for drug repositioning and virtual drug screening. Existing approaches usually transform a 3D protein-ligand complex to a two-dimensional (2D) graph, and then use graph neural networks (GNNs) to predict its binding affinity. However, the node and edge features of the 2D graph are extracted based on invariant local coordinate systems of the 3D complex. As a result, these approaches can not fully learn the global information of the complex, such as the physical symmetry and the topological information of bonds. To address these issues, we propose a novel Equivariant Line Graph Network (ELGN) for binding affinity prediction of 3D protein-ligand complexes. The proposed ELGN firstly adds a super node to the 3D complex, and then builds a line graph based on the 3D complex. After that, ELGN uses a new E(3)-equivariant network layer to pass the messages between nodes and edges based on the global coordinate system of the 3D complex. Experimental results on two real datasets demonstrate the effectiveness of ELGN over several state-of-the-art baselines.

The probability prediction of multivariate time series is a notoriously challenging but practical task. On the one hand, the challenge is how to effectively capture the cross-series correlations between interacting time series, to achieve accurate distribution modeling. On the other hand, we should consider how to capture the contextual information within time series more accurately to model multivariate temporal dynamics of time series. In this work, we proposed a novel non-autoregressive deep learning model, called Multi-scale Attention Normalizing Flow(MANF), where we combine multi-scale attention with relative position information and the multivariate data distribution is represented by the conditioned normalizing flow. Additionally, compared with autoregressive modeling methods, our model avoids the influence of cumulative error and does not increase the time complexity. Extensive experiments demonstrate that our model achieves state-of-the-art performance on many popular multivariate datasets.

Graph contrastive learning (GCL) has attracted a surge of attention due to its superior performance for learning node/graph representations without labels. However, in practice, the underlying class distribution of unlabeled nodes for the given graph is usually imbalanced. This highly imbalanced class distribution inevitably deteriorates the quality of learned node representations in GCL. Indeed, we empirically find that most state-of-the-art GCL methods cannot obtain discriminative representations and exhibit poor performance on imbalanced node classification. Motivated by this observation, we propose a principled GCL framework on Imbalanced node classification (ImGCL), which automatically and adaptively balances the representations learned from GCL without labels. Specifically, we first introduce the online clustering based progressively balanced sampling (PBS) method with theoretical rationale, which balances the training sets based on pseudo-labels obtained from learned representations in GCL. We then develop the node centrality based PBS method to better preserve the intrinsic structure of graphs, by upweighting the important nodes of the given graph. Extensive experiments on multiple imbalanced graph datasets and imbalanced settings demonstrate the effectiveness of our proposed framework, which significantly improves the performance of the recent state-of-the-art GCL methods. Further experimental ablations and analyses show that the ImGCL framework consistently improves the representation quality of nodes in under-represented (tail) classes.

Graph neural networks (GNNs) are popular weapons for modeling relational data. Existing GNNs are not specified for attribute-incomplete graphs, making missing attribute imputation a burning issue. Until recently, many works notice that GNNs are coupled with spectral concentration, which means the spectrum obtained by GNNs concentrates on a local part in spectral domain, e.g., low-frequency due to oversmoothing issue. As a consequence, GNNs may be seriously flawed for reconstructing graph attributes as graph spectral concentration tends to cause a low imputation precision. In this work, we present a regularized graph autoencoder for graph attribute imputation, named MEGAE, which aims at mitigating spectral concentration problem by maximizing the graph spectral entropy. Notably, we first present the method for estimating graph spectral entropy without the eigen-decomposition of Laplacian matrix and provide the theoretical upper error bound. A maximum entropy regularization then acts in the latent space, which directly increases the graph spectral entropy. Extensive experiments show that MEGAE outperforms all the other state-of-the-art imputation methods on a variety of benchmark datasets.

AI-aided drug discovery (AIDD) is gaining popularity due to its potential to make the search for new pharmaceuticals faster, less expensive, and more effective. Despite its extensive use in numerous fields (e.g., ADMET prediction, virtual screening), little research has been conducted on the out-of-distribution (OOD) learning problem with noise. We present DrugOOD, a systematic OOD dataset curator and benchmark for AIDD. Particularly, we focus on the drug-target binding affinity prediction problem, which involves both macromolecule (protein target) and small-molecule (drug compound). DrugOOD offers an automated dataset curator with user-friendly customization scripts, rich domain annotations aligned with biochemistry knowledge, realistic noise level annotations, and rigorous benchmarking of SOTA OOD algorithms, as opposed to only providing fixed datasets. Since the molecular data is often modeled as irregular graphs using graph neural network (GNN) backbones, DrugOOD also serves as a valuable testbed for graph OOD learning problems. Extensive empirical studies have revealed a significant performance gap between in-distribution and out-of-distribution experiments, emphasizing the need for the development of more effective schemes that permit OOD generalization under noise for AIDD.

Proteins govern a wide range of biological systems. Evaluating the changes in protein properties upon protein mutation is a fundamental application of protein design, where modeling the 3D protein structure is a principal task for AI-driven computational approaches. Existing deep learning (DL) approaches represent the protein structure as a 3D geometric graph and simplify the graph modeling to different degrees, thereby failing to capture the low-level atom patterns and high-level amino acid patterns simultaneously. In addition, limited training samples with ground truth labels and protein structures further restrict the effectiveness of DL approaches. In this paper, we propose a new graph learning framework, Hierarchical Graph Invariant Network (HGIN), a fine-grained and data-efficient graph neural encoder for encoding protein structures and predicting the mutation effect on protein properties. For fine-grained modeling, HGIN hierarchically models the low-level interactions of atoms and the high-level interactions of amino acid residues by Graph Neural Networks. For data efficiency, HGIN preserves the invariant encoding for atom permutation and coordinate transformation, which is an intrinsic inductive bias of property prediction that bypasses data augmentations. We integrate HGIN into a Siamese network to predict the quantitative effect on protein properties upon mutations. Our approach outperforms 9 state-of-the-art approaches on 3 protein datasets. More inspiringly, when predicting the neutralizing ability of human antibodies against COVID-19 mutant viruses, HGIN achieves an absolute improvement of 0.23 regarding the Spearman coefficient.

By simply taking convex combinations between pairs of samples and their labels, mixup training has been shown to easily improve predictive accuracy. It has been recently found that models trained with mixup also perform well on uncertainty calibration. However, in this study, we found that mixup training usually makes models less calibratable than vanilla empirical risk minimization, which means that it would harm uncertainty estimation when post-hoc calibration is considered. By decomposing the mixup process into data transformation and random perturbation, we suggest that the confidence penalty nature of the data transformation is the reason of calibration degradation. To mitigate this problem, we first investigate the mixup inference strategy and found that despite it improves calibration on mixup, this ensemble-like strategy does not necessarily out-perform simple ensemble. Then, we propose a general strategy named mixup inference in training, which adopts a simple decoupling principle for recovering the outputs of raw samples at the end of forward network pass. By embedding the mixup inference, models can be learned from the original one-hot labels and hence avoid the negative impact of confidence penalty. Our experiments show this strategy properly solves mixup's calibration issue without sacrificing the predictive performance, while even improves accuracy than vanilla mixup.

It's widely acknowledged that deep learning models with flatter minima in its loss landscape tend to generalize better. However, such property is under-explored in deep long-tailed recognition (DLTR), a practical problem where the model is required to generalize equally well across all classes when trained on highly imbalanced label distribution. In this paper, through empirical observations, we argue that sharp minima are in fact prevalent in deep long-tailed models, whereas naï ve integration of existing flattening operations into long-tailed learning algorithms brings little improvement. Instead, we propose an effective two-stage sharpness-aware optimization approach based on the decoupling paradigm in DLTR. In the first stage, both the feature extractor and classifier are trained under parameter perturbations at a class-conditioned scale, which is theoretically motivated by the characteristic radius of flat minima under the PAC-Bayesian framework. In the second stage, we generate adversarial features with class-balanced sampling to further robustify the classifier with the backbone frozen. Extensive experiments on multiple long-tailed visual recognition benchmarks show that, our proposed Class-Conditional Sharpness-Aware Minimization (CC-SAM), achieves competitive performance compared to the state-of-the-arts. Code is available at https://github.com/zzpustc/CC-SAM.

Portfolio Selection is an important real-world financial task and has attracted extensive attention in artificial intelligence communities. This task, however, has two main difficulties: (i) the non-stationary price series and complex asset correlations make the learning of feature representation very hard; (ii) the practicality principle in financial markets requires controlling both transaction and risk costs. Most existing methods adopt handcraft features and/or consider no constraints for the costs, which may make them perform unsatisfactorily and fail to control both costs in practice. In this paper, we propose a cost-sensitive portfolio selection method with deep reinforcement learning. Specifically, a novel two-stream portfolio policy network is devised to extract both price series patterns and asset correlations, while a new cost-sensitive reward function is developed to maximize the accumulated return and constrain both costs via reinforcement learning. We theoretically analyze the near-optimality of the proposed reward, which shows that the growth rate of the policy regarding this reward function can approach the theoretical optimum. We also empirically evaluate the proposed method on real-world datasets. Promising results demonstrate the effectiveness and superiority of the proposed method in terms of profitability, cost-sensitivity and representation abilities.

The exploitation and exploration dilemma is a crucial issue in bandit online multiclass prediction. Conventional algorithms typically resort to either random sample or estimate uncertainty for exploration. In contrast, we propose a novel scheme that focuses solely on exploitation with implicit feedback. To ensure efficient information feedback even when predictions are incorrect, we introduce mixed losses into our proposed scheme. We derive two mixed versions of the multiclass hinge loss and the logistic loss, along with their corresponding algorithms. Context-free and context-aware experiments are conducted to evaluate the performance of our proposed algorithms. Experiment results demonstrate that random sampling are unnecessary if a reasonable loss function is employed. By compared with several state-of-the-art baselines on both synthetic and real-world datasets, our proposed algorithms show their superior performance. Remarkably, they even outperform the Perceptron algorithm with full-information feedback in some cases.

Retrosynthetic planning aims to devise a complete multi-step synthetic route from starting materials to a target molecule. Current strategies use a decoupled approach of single-step retrosynthesis models and search algorithms, taking only the product as the input to predict the reactants for each planning step and ignoring valuable context information along the synthetic route. In this work, we propose a novel framework that utilizes context information for improved retrosynthetic planning. We view synthetic routes as reaction graphs and propose to incorporate context through three principled steps: encode molecules into embeddings, aggregate information over routes, and readout to predict reactants. Our approach is the first attempt to utilize in-context learning for retrosynthesis prediction in retrosynthetic planning. The entire framework can be efficiently optimized in an end-to-end fashion and produce more practical and accurate predictions. Comprehensive experiments demonstrate that by fusing in the context information over routes, our model significantly improves the performance of retrosynthetic planning over baselines that are not context-aware, especially for long synthetic routes. Code is available at https://github.com/SongtaoLiu0823/FusionRetro.

Organic reaction prediction is a critical task in drug discovery. Recently, researchers have achieved non-autoregressive reaction prediction by modeling the redistribution of electrons, resulting in state-of-the-art top-1 accuracy, and enabling parallel sampling. However, the current non-autoregressive decoder does not satisfy two essential rules of electron redistribution modeling simultaneously: the electron-counting rule and the symmetry rule. This violation of the physical constraints of chemical reactions impairs model performance. In this work, we propose a new framework called ReactionSink that combines two doubly stochastic self-attention mappings to obtain electron redistribution predictions that follow both constraints. We further extend our solution to a general multi-head attention mechanism with augmented constraints. To achieve this, we apply Sinkhorn's algorithm to iteratively update self-attention mappings, which imposes doubly conservative constraints as additional informative priors on electron redistribution modeling. We theoretically demonstrate that our ReactionSink can simultaneously satisfy both rules, which the current decoder mechanism cannot do. Empirical results show that our approach consistently improves the predictive performance of non-autoregressive models and does not bring an unbearable additional computational cost.

Reaction and retrosynthesis prediction are two fundamental tasks in computational chemistry. In recent years, these two tasks have attracted great attentions from both machine learning and drug discovery communities. Various deep learning approaches have been proposed to tackle these two problems and achieved initial success. In this survey, we conduct a comprehensive investigation on advanced deep learning-based reaction and retrosynthesis prediction models. We first summarize the design mechanism, strengths and weaknesses of the state-of-the-art approaches. Then we further discuss limitations of current solutions and open challenges in the problem itself. Last but not the least, we present some promising directions to facilitate future research. To our best knowledge, this paper is the first comprehensive and systematic survey on unified understanding of reaction and retrosynthesis prediction.

Recent awareness of privacy protection and compliance requirement resulted in a controversial view of recommendation system due to personal data usage. Therefore, privacy-protected recommendation emerges as a novel research direction. In this paper, we first formulate this problem as a vertical federated learning problem, i.e., features are vertically distributed over different departments. We study a contextual bandit learning problem for recommendation in the vertical federated setting. To this end, we carefully design a customized encryption scheme named orthogonal matrix-based mask mechanism (O3M). O3M mechanism, a tailored component for contextual bandits by carefully exploiting their shared structure, can ensure privacy protection while avoiding expensive conventional cryptographic techniques. We further apply the mechanism to two commonly-used bandit algorithms, LinUCB and LinTS, and instantiate two practical protocols for online recommendation. The proposed protocols can perfectly recover the service quality of centralized bandit algorithms while achieving a satisfactory runtime efficiency, which is theoretically proved and analysed in this paper. By conducting extensive experiments on both synthetic and real-world datasets, we show the superiority of the proposed method in terms of privacy protection and recommendation performance.

The performance of deep learning models heavily relies on the quality of the training data. Inadequacies in the training data, such as corrupt input or noisy labels, can lead to the failure of model generalization. Recent studies propose repairing the model by identifying the training samples that contribute to the failure and removing their influence from the model. However, it is important to note that the identified data may contain both beneficial and detrimental information. Simply erasing the information of the identified data from the model can have a negative impact on its performance, especially when accurate data is mistakenly identified as detrimental and removed. To overcome this challenge, we propose a novel approach that leverages the knowledge obtained from a retained clean set. Our method first identifies harmful data by utilizing the clean set, then separates the beneficial and detrimental information within the identified data. Finally, we utilize the extracted beneficial information to enhance the model's performance. Through empirical evaluations, we demonstrate that our method outperforms baseline approaches in both identifying harmful data and rectifying model failures. Particularly in scenarios where identification is challenging and a significant amount of benign data is involved, our method improves performance while the baselines deteriorate due to the erroneous removal of beneficial information.

Large language models have demonstrated surprising ability to perform in-context learning, i.e., these models can be directly applied to solve numerous downstream tasks by conditioning on a prompt constructed by a few input-output examples. However, prior research has shown that in-context learning can suffer from high instability due to variations in training examples, example order, and prompt formats. Therefore, the construction of an appropriate prompt is essential for improving the performance of in-context learning. In this paper, we revisit this problem from the view of predictive bias. Specifically, we introduce a metric to evaluate the predictive bias of a fixed prompt against labels or a given attributes. Then we empirically show that prompts with higher bias always lead to unsatisfactory predictive quality. Based on this observation, we propose a novel search strategy based on the greedy search to identify the near-optimal prompt for improving the performance of in-context learning. We perform comprehensive experiments with state-of-the-art mainstream models such as GPT-3 on various downstream tasks. Our results indicate that our method can enhance the model's in-context learning performance in an effective and interpretable manner.

With a long history of traditional Graph Anomaly Detection (GAD) algorithms and recently popular Graph Neural Networks (GNNs), it is still not clear (1) how they perform under a standard comprehensive setting, (2) whether GNNs can outperform traditional algorithms such as tree ensembles, and (3) how about their efficiency on large-scale graphs. In response, we introduce GADBench---a benchmark tool dedicated to supervised anomalous node detection in static graphs. GADBench facilitates a detailed comparison across 29 distinct models on ten real-world GAD datasets, encompassing thousands to millions (~6M) nodes. Our main finding is that tree ensembles with simple neighborhood aggregation can outperform the latest GNNs tailored for the GAD task. We shed light on the current progress of GAD, setting a robust groundwork for subsequent investigations in this domain. GADBench is open-sourced at https://github.com/squareRoot3/GADBench.

Molecule generation plays an important role in accelerating drug discovery. In recent years, many molecule generation methods have been proposed based on variational autoencoders (VAEs), due to its advantages in latent manifold representation learning and training stability. However, most of the existing VAE-based models require tedious graph matching operations during training, and tend to generate invalid molecules. To overcome these limitations, in this paper, we propose a novel molecular graph variational autoencoder (MoVAE). Firstly, to avoid complicated graph matching, the proposed MoVAE only encodes and decodes all the nodes and edges individually. Secondly, to improve the generation validity, it adversarially trains the model by treating the encoder and decoder as the discriminator and generator. In addition, to generate molecules with various target conditions, the MoVAE also introduces drug property constraints and valence histogram constraints. Experiment results on two real datasets show that our model outperforms almost all the state-of-the-art algorithms.

Predicting molecular properties is known to be a difficult few-shot learning problem in drug discovery due to high failure rates of virtual screening. Meta-learning frameworks have been successfully leveraged to tackle this few-shot molecular property prediction problem. Recent work has shown that the generalization of meta- learning can be further improved by task augmentation techniques. However, it is challenging to directly apply current popular task augmentation strategies to molecular modeling due to its highly discrete nature. To this end, we propose a motif-based task augmentation (MTA) technique to address this challenge. MTA is mainly conducting task augmentations by generating new labeled samples through retrieving highly relevant motifs from a pre-defined motif vocabulary as an external memory. Both support samples and query samples are interpolated with their corresponding retrieved motifs in the latent space in a convex manner parameterized by a specific probability distribution. Augmented query samples are classified by measuring their Euclidean distances to augmented prototype clusters. With our novel MTA approach, the above two overfitting issues can be alleviated through establishing connections between different tasks with augmented motifs. Empirical results for popular benchmark datasets demonstrate that our approach can consistently outperform existing baseline methods and can also provide some interpretable relation information between structural motifs and molecules.

Despite the success of graph neural networks (GNNs) over the Web in recent years, the typical transductive learning setting for node classification requires GNNs to be retrained frequently, making them vulnerable to poisoning attacks by corrupting the training graph. Poisoning attacks on graphs are, however, non-trivial as the attack space is potentially large, and the discrete graph structure makes the poisoning function non-differentiable. In this paper, we revisit the bi-level optimization problem in graph poisoning and propose a novel graph poisoning method, termed Curriculum Graph Poisoning (CuGPo), inspired by curriculum learning. In contrast to other poisoning attacks that use heuristics or directly optimize the graph, our method learns to generate poisoned graphs from basic adversarial knowledge first and advanced knowledge later. Specifically, for the outer optimization, we utilize the slightly perturbed graphs which represent the easy poisoning task at the beginning, and then enlarge the attack space until the final; for the inner optimization, we firstly exploit the knowledge from the clean graph and then adapt quickly to perturbed graphs to obtain the adversarial knowledge. Extensive experiments demonstrate that CuGPo achieves state-of-the-art performance in graph poisoning attacks.

Neural-symbolic learning, aiming to combine the perceiving power of neural perception and the reasoning power of symbolic logic together, has drawn increasing research attention. However, existing works simply cascade the two components together and optimize them isolatedly, failing to utilize the mutual enhancing information between them. To address this problem, we propose DeepLogic, a framework with joint learning of neural perception and logical reasoning, such that these two components are jointly optimized through mutual supervision signals. In particular, the proposed DeepLogic framework contains a deep-logic module that is capable of representing complex first-order-logic formulas in a tree structure with basic logic operators. We then theoretically quantify the mutual supervision signals and propose the deep&logic optimization algorithm for joint optimization. We further prove the convergence of DeepLogic and conduct extensive experiments on model performance, convergence, and generalization, as well as its extension to the continuous domain. The experimental results show that through jointly learning both perceptual ability and logic formulas in a weakly supervised manner, our proposed DeepLogic framework can significantly outperform DNN-based baselines by a great margin and beat other strong baselines without out-of-box tools.

Decentralized online learning (online learning in decentralized networks) has been attracting more and more attention, since it is believed that decentralized online learning can help data providers cooperatively better solve their online problems without sharing their private data to a third party or other providers. Typically, the cooperation is achieved by letting the data providers exchange their models between neighbors, e.g., recommendation model. However, the best regret bound for a decentralized online learning algorithm is 𝒪( n √ T ), where n is the number of nodes (or users) and T is the number of iterations. This is clearly insignificant, since this bound can be achieved without any communication in the networks. This reminds us to ask a fundamental question: Can people really get benefit from the decentralized online learning by exchanging information? In this article, we studied when and why the communication can help the decentralized online learning to reduce the regret. Specifically, each loss function is characterized by two components: the adversarial component and the stochastic component. Under this characterization, we show that decentralized online gradient enjoys a regret bound \( {\mathcal {O}(\sqrt {n^2TG^2 + n T \sigma ^2})} \) , where G measures the magnitude of the adversarial component in the private data (or equivalently the local loss function) and σ measures the randomness within the private data. This regret suggests that people can get benefits from the randomness in the private data by exchanging private information. Another important contribution of this article is to consider the dynamic regret—a more practical regret to track users’ interest dynamics. Empirical studies are also conducted to validate our analysis.

Retrosynthetic planning plays an important role in the field of organic chemistry, which could generate a synthetic route for the target product. The synthetic route is a series of reactions which are started from the available molecules. The most challenging problem in the generation of the synthetic route is the large search space of the candidate reactions. Estimating the cost of candidate reactions has been proved effectively to prune the search space, which could achieve a higher accuracy with the same search iteration. And the estimation of one reaction is comprised of the estimations of all its reactants. So, how to estimate the cost of these reactants will directly influence the quality of results. To get a better performance, we propose a new framework, named GNN-Retro, for retrosynthetic planning problem by combining graph neural networks(GNN) and the latest search algorithm. The structure of GNN in our framework could incorporate the information of neighboring molecules, which will improve the estimation accuracy of our framework. The experiments on the USPTO dataset show that our framework could outperform the state-of-the-art methods with a large margin under the same settings.

Inferring gene regulatory networks based on single-cell transcriptomes is critical for systematically understanding cell-specific regulatory networks and discovering drug targets in tumor cells. Here we show that existing methods mainly perform co-expression analysis and apply the image-based model to deal with the non-euclidean scRNA-seq data, which may not reasonably handle the dropout problem and not fully take advantage of the validated gene regulatory topology. We propose a graph-based end-to-end deep learning model for GRN inference (GRNInfer) with the help of known regulatory relations through transductive learning. The robustness and superiority of the model are demonstrated by comparative experiments.

In this paper, we propose a novel sequence verification task that aims to distinguish positive video pairs performing the same action sequence from negative ones with step-level transformations but still conducting the same task. Such a challenging task resides in an open-set setting without prior action detection or segmentation that requires event-level or even frame-level annotations. To that end, we carefully reorganize two publicly available action-related datasets with step-procedure-task structure. To fully investigate the effectiveness of any method, we collect a scripted video dataset enumerating all kinds of step-level transformations in chemical experiments. Besides, a novel evaluation metric Weighted Distance Ratio is introduced to ensure equivalence for different step-level transformations during evaluation. In the end, a simple but effective baseline based on the transformer encoder with a novel sequence alignment loss is introduced to better characterize long-term dependency between steps, which outperforms other action recognition methods. Codes and data will be released1:

Graph neural networks (GNNs) have demonstrated superior performance for semi-supervised node classification on graphs, as a result of their ability to exploit node features and topological information simultaneously. However, most GNNs implicitly assume that the labels of nodes and their neighbors in a graph are the same or consistent, which does not hold in heterophilic graphs, where the labels of linked nodes are likely to differ. Moreover, when the topology is non-informative for label prediction, ordinary GNNs may work significantly worse than simply applying multi-layer perceptrons (MLPs) on each node. To tackle the above problem, we propose a new $p$-Laplacian based GNN model, termed as $^p$GNN, whose message passing mechanism is derived from a discrete regularization framework and could be theoretically explained as an approximation of a polynomial graph filter defined on the spectral domain of $p$-Laplacians. The spectral analysis shows that the new message passing mechanism works as low-high-pass filters, thus making $^p$GNNs are effective on both homophilic and heterophilic graphs. Empirical studies on real-world and synthetic datasets validate our findings and demonstrate that $^p$GNNs significantly outperform several state-of-the-art GNN architectures on heterophilic benchmarks while achieving competitive performance on homophilic benchmarks. Moreover, $^p$GNNs can adaptively learn aggregation weights and are robust to noisy edges.

Graph Neural Networks (GNNs) have achieved remarkable performance on graph-based tasks. The key idea for GNNs is to obtain informative representation through aggregating information from local neighborhoods. However, it remains an open question whether the neighborhood information is adequately aggregated for learning representations of nodes with few neighbors. To address this, we propose a simple and efficient data augmentation strategy, local augmentation, to learn the distribution of the node representations of the neighbors conditioned on the central node’s representation and enhance GNN’s expressive power with generated features. Local augmentation is a general framework that can be applied to any GNN model in a plug-and-play manner. It samples feature vectors associated with each node from the learned conditional distribution as additional input for the backbone model at each training iteration. Extensive experiments and analyses show that local augmentation consistently yields performance improvement when applied to various GNN architectures across a diverse set of benchmarks. For example, experiments show that plugging in local augmentation to GCN and GAT improves by an average of 3.4% and 1.6% in terms of test accuracy on Cora, Citeseer, and Pubmed. Besides, our experimental results on large graphs (OGB) show that our model consistently improves performance over backbones. Code is available at https://github.com/SongtaoLiu0823/LAGNN.

Test-time adaptation provides an effective means of tackling the potential distribution shift between model training and inference, by dynamically updating the model at test time. This area has seen fast progress recently, at the effectiveness of handling test shifts. Nonetheless, prior methods still suffer two key limitations: 1) these methods rely on performing backward computation for each test sample, which takes a considerable amount of time; and 2) these methods focus on improving the performance on out-of-distribution test samples and ignore that the adaptation on test data may result in a catastrophic forgetting issue, \ie, the performance on in-distribution test samples may degrade. To address these issues, we propose an efficient anti-forgetting test-time adaptation (EATA) method. Specifically, we devise a sample-efficient entropy minimization loss to exclude uninformative samples out of backward computation, which improves the overall efficiency and meanwhile boosts the out-of-distribution accuracy. Afterward, we introduce a regularization loss to ensure that critical model weights tend to be preserved during adaptation, thereby alleviating the forgetting issue. Extensive experiments on CIFAR-10-C, ImageNet-C, and ImageNet-R verify the effectiveness and superiority of our EATA.

Subpopulation shift widely exists in many real-world machine learning applications, referring to the training and test distributions containing the same subpopulation groups but varying in subpopulation frequencies. Importance reweighting is a normal way to handle the subpopulation shift issue by imposing constant or adaptive sampling weights on each sample in the training dataset. However, some recent studies have recognized that most of these approaches fail to improve the performance over empirical risk minimization especially when applied to over-parameterized neural networks. In this work, we propose a simple yet practical framework, called uncertainty-aware mixup (UMIX), to mitigate the overfitting issue in over-parameterized models by reweighting the ''mixed'' samples according to the sample uncertainty. The training-trajectories-based uncertainty estimation is equipped in the proposed UMIX for each sample to flexibly characterize the subpopulation distribution. We also provide insightful theoretical analysis to verify that UMIX achieves better generalization bounds over prior works. Further, we conduct extensive empirical studies across a wide range of tasks to validate the effectiveness of our method both qualitatively and quantitatively. Code is available at https://github.com/TencentAILabHealthcare/UMIX.

Learning set functions becomes increasingly important in many applications like product recommendation and compound selection in AI-aided drug discovery. The majority of existing works study methodologies of set function learning under the function value oracle, which, however, requires expensive supervision signals. This renders it impractical for applications with only weak supervisions under the Optimal Subset (OS) oracle, the study of which is surprisingly overlooked. In this work, we present a principled yet practical maximum likelihood learning framework, termed as EquiVSet, that simultaneously meets the following desiderata of learning neural set functions under the OS oracle: i) permutation invariance of the set mass function being modeled; ii) permission of varying ground set; iii) minimum prior and iv) scalability. The main components of our framework involve: an energy-based treatment of the set mass function, DeepSet-style architectures to handle permutation invariance, mean-field variational inference, and its amortized variants. Thanks to the delicate combination of these advanced architectures, empirical studies on three real-world applications (including Amazon product recommendation, set anomaly detection, and compound selection for virtual screening) demonstrate that EquiVSet outperforms the baselines by a large margin.

Scaling reinforcement learning (RL) to recommender systems (RS) is promising since maximizing the expected cumulative rewards for RL agents meets the objective of RS, i.e., improving customers' long-term satisfaction. A key approach to this goal is offline RL, which aims to learn policies from logged data rather than expensive online interactions. In this paper, we propose Value Penalized Q-learning (VPQ), a novel uncertainty-based offline RL algorithm that penalizes the unstable Q-values in the regression target using uncertainty-aware weights, achieving the conservative Q-function without the need of estimating the behavior policy, suitable for RS with a large number of items. Experiments on two real-world datasets show the proposed method serves as a gain plug-in for existing RS models.

Click-Through Rate (CTR) prediction, which aims to estimate the probability that a user will click an item, is an essential component of online advertising. Existing methods mainly attempt to mine user interests from users' historical behaviours, which contain users' directly interacted items. Although these methods have made great progress, they are often limited by the recommender system's direct exposure and inactive interactions, and thus fail to mine all potential user interests. To tackle these problems, we propose Neighbor-Interaction based CTR prediction (NI-CTR), which considers this task under a Heterogeneous Information Network (HIN) setting. In short, Neighbor-Interaction based CTR prediction involves the local neighborhood of the target user-item pair in the HIN to predict their linkage. In order to guide the representation learning of the local neighbourhood, we further consider different kinds of interactions among the local neighborhood nodes from both explicit and implicit perspective, and propose a novel Graph-Masked Transformer (GMT) to effectively incorporates these kinds of interactions to produce highly representative embeddings for the target user-item pair. Moreover, in order to improve model robustness against neighbour sampling, we enforce a consistency regularization loss over the neighbourhood embedding. We conduct extensive experiments on two real-world datasets with millions of instances and the experimental results show that our proposed method outperforms state-of-the-art CTR models significantly. Meanwhile, the comprehensive ablation studies verify the effectiveness of every component of our model. Furthermore, we have deployed this framework on the WeChat Official Account Platform with billions of users. The online A/B tests demonstrate an average CTR improvement of 21.9% against all online baselines.

Fake News detection has attracted much attention in recent years. Social context based detection methods attempt to model the spreading patterns of fake news by utilizing the collective wisdom from users on social media. This task is challenging for three reasons: (1) There are multiple types of entities and relations in social context, requiring methods to effectively model the heterogeneity. (2) The emergence of news in novel topics in social media causes distribution shifts, which can significantly degrade the performance of fake news detectors. (3) Existing fake news datasets usually lack of great scale, topic diversity and user social relations, impeding the development of this field. To solve these problems, we formulate social context based fake news detection as a heterogeneous graph classification problem, and propose a fake news detection model named Post-User Interaction Network (PSIN), which adopts a divide-and-conquer strategy to model the post-post, user-user and post-user interactions in social context effectively while maintaining their intrinsic characteristics. Moreover,we adopt an adversarial topic discriminator for topic-agnostic feature learning, in order to improve the generalizability of our method for new-emerging topics. Furthermore, we curate a new dataset for fake news detection, which contains over 27,155 news from 5 topics, 5 million posts, 2 million users and their induced social graph with 0.2 billion edges. It has been published on https://github.com/qwerfdsaplking/MC-Fake. Extensive experiments illustrate that our method outperforms SOTA baselines in both in-topic and out-of-topic settings.

Designing effective architectures is one of the key factors behind the success of deep neural networks. Existing deep architectures are either manually designed or automatically searched by some Neural Architecture Search (NAS) methods. However, even a well-designed/searched architecture may still contain many nonsignificant or redundant modules/operations (e.g., some intermediate convolution or pooling layers). Such redundancy may not only incur substantial memory consumption and computational cost but also deteriorate the performance. Thus, it is necessary to optimize the operations inside an architecture to improve the performance without introducing extra computational cost. To this end, we have proposed a Neural Architecture Transformer (NAT) method which casts the optimization problem into a Markov Decision Process (MDP) and seeks to replace the redundant operations with more efficient operations, such as skip or null connection. Note that NAT only considers a small number of possible replacements/transitions and thus comes with a limited search space. As a result, such a small search space may hamper the performance of architecture optimization. To address this issue, we propose a Neural Architecture Transformer++ (NAT++) method which further enlarges the set of candidate transitions to improve the performance of architecture optimization. Specifically, we present a two-level transition rule to obtain valid transitions, i.e., allowing operations to have more efficient types (e.g., convolution → separable convolution) or smaller kernel sizes (e.g., 5×5 → 3×3). Note that different operations may have different valid transitions. We further propose a Binary-Masked Softmax (BMSoftmax) layer to omit the possible invalid transitions. Last, based on the MDP formulation, we apply policy gradient to learn an optimal policy, which will be used to infer the optimized architectures. Extensive experiments show that the transformed architectures significantly outperform both their original counterparts and the architectures optimized by existing methods.

Temporal action localization, which requires a machine to recognize the location as well as the category of action instances in videos, has long been researched in computer vision. The main challenge of temporal action localization lies in that videos are usually long and untrimmed with diverse action contents involved. Existing state-of-the-art action localization methods divide each video into multiple action units (i.e., proposals in two-stage methods and segments in one-stage methods) and then perform action recognition/regression on each of them individually, without explicitly exploiting their relations during learning. In this paper, we claim that the relations between action units play an important role in action localization, and a more powerful action detector should not only capture the local content of each action unit but also allow a wider field of view on the context related to it. To this end, we propose a general graph convolutional module (GCM) that can be easily plugged into existing action localization methods, including two-stage and one-stage paradigms. To be specific, we first construct a graph, where each action unit is represented as a node and their relations between two action units as an edge. Here, we use two types of relations, one for capturing the temporal connections between different action units, and the other one for characterizing their semantic relationship. Particularly for the temporal connections in two-stage methods, we further explore two different kinds of edges, one connecting the overlapping action units and the other one connecting surrounding but disjointed units. Upon the graph we built, we then apply graph convolutional networks (GCNs) to model the relations among different action units, which is able to learn more informative representations to enhance action localization. Experimental results show that our GCM consistently improves the performance of existing action localization methods, including two-stage methods (e.g., CBR [15] and R-C3D [47]) and one-stage methods (e.g., D-SSAD [22]), verifying the generality and effectiveness of our GCM. Moreover, with the aid of GCM, our approach significantly outperforms the state-of-the-art on THUMOS14 (50.9 percent versus 42.8 percent). Augmentation experiments on ActivityNet also verify the efficacy of modeling the relationships between action units. The source code and the pre-trained models are available at https://github.com/Alvin-Zeng/GCM.

The task of point of interest (POI) recommendation aims to recommend unvisited places to users based on their check-in history. A major challenge in POI recommendation is data sparsity, because a user typically visits only a very small number of POIs among all available POIs. In this paper, we propose AUC-MF to address the POI recommendation problem by maximizing Area Under the ROC curve (AUC). AUC has been widely used for measuring classification performance with imbalanced data distributions. To optimize AUC, we transform the recommendation task to a classification problem, where the visited locations are positive examples and the unvisited are negative ones. We define a new lambda for AUC to utilize the LambdaMF model, which combines the lambda-based method and matrix factorization model in collaborative filtering. Many studies have shown that geographic information plays an important role in POI recommendation. In this study, we focus on two levels geographic information: local similarity and global similarity. We further show that AUC-MF can be easily extended to incorporate geographical contextual information for POI recommendation. Specifically, we propose two novel methods to incorporate geographical information in AUC-MF. Different from most existing models where the contextual information are incorporated into the objective function, the incorporation of contextual information in AUC-MF is a refinement of the model and a sampling strategy. The sampling strategy could speedup convergence and the refining of recommendations is independent of training of the model. This mechanism also enables AUC-MF to be able produce recommendations refined towards different contextual information, with minimum computational cost. Experiments on two datasets show that the proposed AUC-MF outperforms state-of-the-art methods significantly in terms of recommendation accuracy.

Portfolio Selection is an important real-world financial task and has attracted extensive attention in artificial intelligence communities. This task, however, has two main difficulties: (i) the non-stationary price series and complex asset correlations make the learning of feature representation very hard; (ii) the practicality principle in financial markets requires controlling both transaction and risk costs. Most existing methods adopt handcraft features and/or consider no constraints for the costs, which may make them perform unsatisfactorily and fail to control both costs in practice. In this paper, we propose a cost-sensitive portfolio selection method with deep reinforcement learning. Specifically, a novel two-stream portfolio policy network is devised to extract both price series patterns and asset correlations, while a new cost-sensitive reward function is developed to maximize the accumulated return and constrain both costs via reinforcement learning. We theoretically analyze the near-optimality of the proposed reward, which shows that the growth rate of the policy regarding this reward function can approach the theoretical optimum. We also empirically evaluate the proposed method on real-world datasets. Promising results demonstrate the effectiveness and superiority of the proposed method in terms of profitability, cost-sensitivity and representation abilities.

We present JueWu-SL, the first supervised-learning-based artificial intelligence (AI) program that achieves human-level performance in playing multiplayer online battle arena (MOBA) games. Unlike prior attempts, we integrate the macro-strategy and the micromanagement of MOBA-game-playing into neural networks in a supervised and end-to-end manner. Tested on Honor of Kings, the most popular MOBA at present, our AI performs competitively at the level of High King players in standard 5v5 games.

Protein secondary structure prediction (PSSP) is an essential task in computational biology. To achieve the accurate PSSP, the general and vital feature engineering is to use multiple sequence alignment (MSA) for Position-Specific Scoring Matrix (PSSM) extraction. However, when only low-quality PSSM can be obtained due to poor sequence homology, previous PSSP accuracy (merely around 65%) is far from practical usage for subsequent tasks. In this paper, we propose a novel PSSM-Distil framework for PSSP on low-quality PSSM, which not only enhances the PSSM feature at a lower level but also aligns the feature distribution at a higher level. In practice, the PSSM-Distil first exploits the proteins with high-quality PSSM to achieve a teacher network for PSSP in a full-supervised way. Under the guidance of the teacher network, the low-quality PSSM and corresponding student network with low discriminating capacity are effectively resolved by feature enhancement through EnhanceNet and distribution alignment through knowledge distillation with contrastive learning. Further, our PSSM-Distil supports the input from a pre-trained protein sequence language BERT model to provide auxiliary information, which is designed to address the extremely low-quality PSSM cases, i.e., no homologous sequence. Extensive experiments demonstrate the proposed PSSM-Distil outperforms state-of-the-art models on PSSP by 6% on average and nearly 8% in extremely low-quality cases on public benchmarks, BC40 and CB513.

Graph Neural Networks (GNNs) draw their strength from explicitly modeling the topological information of structured data. However, existing GNNs suffer from limited capability in capturing the hierarchical graph representation which plays an important role in graph classification. In this paper, we innovatively propose hierarchical graph capsule network (HGCN) that can jointly learn node embeddings and extract graph hierarchies. Specifically, disentangled graph capsules are established by identifying heterogeneous factors underlying each node, such that their instantiation parameters represent different properties of the same entity. To learn the hierarchical representation, HGCN characterizes the part-whole relationship between lower-level capsules (part) and higher-level capsules (whole) by explicitly considering the structure information among the parts. Experimental studies demonstrate the effectiveness of HGCN and the contribution of each component. Code: https://github.com/uta-smile/HGCN

Decentralized optimization problems frequently appear in the large scale machine learning problems. However, few works work on the difficult nonconvex nonsmooth case. In this paper, we propose a decentralized primal-dual algorithm to solve this type of problem in a decentralized manner and the proposed algorithm can achieve an $\mathcal{O}(1/\epsilon^2)$ iteration complexity to attain an $\epsilon-$solution, which is the well-known lower iteration complexity bound for nonconvex optimization. To our knowledge, it is the first algorithm achieving this rate under a nonconvex, nonsmooth decentralized setting. Furthermore, to reduce communication overhead, we also modifying our algorithm by compressing the vectors exchanged between agents. The iteration complexity of the algorithm with compression is still $\mathcal{O}(1/\epsilon^2)$. Besides, we apply the proposed algorithm to solve nonconvex linear regression problem and train deep learning model, both of which demonstrate the efficiency and efficacy of the proposed algorithm.

Recent studies imply that deep neural networks are vulnerable to adversarial examples, i.e., inputs with a slight but intentional perturbation are incorrectly classified by the network. Such vulnerability makes it risky for some security-related applications (e.g., semantic segmentation in autonomous cars) and triggers tremendous concerns on the model reliability. For the first time, we comprehensively evaluate the robustness of existing UDA methods and propose a robust UDA approach. It is rooted in two observations: i) the robustness of UDA methods in semantic segmentation remains unexplored, which poses a security concern in this field; and ii) although commonly used self-supervision (e.g., rotation and jigsaw) benefits model robustness in classification and recognition tasks, they fail to provide the critical supervision signals that are essential in semantic segmentation. These observations motivate us to propose adversarial self-supervision UDA (or ASSUDA) that maximizes the agreement between clean images and their adversarial examples by a contrastive loss in the output space. Extensive empirical studies on commonly used benchmarks demonstrate that ASSUDA is resistant to adversarial attacks.

In real-world applications, data often come in a growing manner, where the data volume and the number of classes may increase dynamically. This will bring a critical challenge for learning: given the increasing data volume or the number of classes, one has to instantaneously adjust the neural model capacity to obtain promising performance. Existing methods either ignore the growing nature of data or seek to independently search an optimal architecture for a given dataset, and thus are incapable of promptly adjusting the architectures for the changed data. To address this, we present a neural architecture adaptation method, namely Adaptation eXpert (AdaXpert), to efficiently adjust previous architectures on the growing data. Specifically, we introduce an architecture adjuster to generate a suitable architecture for each data snapshot, based on the previous architecture and the different extent between current and previous data distributions. Furthermore, we propose an adaptation condition to determine the necessity of adjustment, thereby avoiding unnecessary and time-consuming adjustments. Extensive experiments on two growth scenarios (increasing data volume and number of classes) demonstrate the effectiveness of the proposed method.

The surrogate that predicts the performance of hyperparameters has been a key component for sequential model-based hyperparameter optimization. In practical applications, a trial of a hyper-parameter configuration may be so costly that a surrogate is expected to return an optimal configuration with as few trials as possible. Observing that human experts draw on their expertise in a machine learning model by trying configurations that once performed well on other datasets, we are inspired to build a trial-efficient surrogate by transferring the meta-knowledge learned from historical trials on other datasets. We propose an end-to-end surrogate named as Transfer NeuralProcesses (TNP) that learns a comprehensive set of meta-knowledge, including the parameters of historical surrogates, historical trials, and initial configurations for other datasets. Experiments on extensive OpenML datasets and three computer vision datasets demonstrate that the proposed algorithm achieves state-of-the-art performance in at least one order of magnitude less trials.

Mathematical expression recognition (MER) aims to convert an image of mathematical expressions into a Latex sequence. In practice, the task of MER is challenging, since 1) the images of mathematical expressions often contain complex structure relationships, e.g., fractions, matrixes, and subscripts; 2) the generated Latex sequences can be very complex and they have to satisfy strict syntax rules. Existing methods, however, often ignore the complex dependence among image regions, resulting in poor feature representation. In addition, they may fail to capture the rigorous relations among different formula symbols as they consider MER as a common language generation task. To address these issues, we propose a Structure-Aware Sequence-Level (SASL) model for MER. First, to better represent and recognize the visual content of formula images, we propose a structure-aware module to capture the relationship among different symbols. Meanwhile, the sequence-level modeling helps the model to concentrate on the generation of entire sequences. To make the problem feasible, we cast the generation problem into a Markov decision process (MDP) and seek to learn a Latex sequence generating policy. Based on MDP, we learn SASL by maximizing the matching score of each image-sequence pair to obtain the generation policy. Extensive experiments on the IM2LATEX-100K dataset verify the effectiveness and superiority of the proposed method.

To benefit the learning of a new task, meta-learning has been proposed to transfer a well-generalized meta-model learned from various meta-training tasks. Existing meta-learning algorithms randomly sample meta-training tasks with a uniform probability, under the assumption that tasks are of equal importance. However, it is likely that tasks are detrimental with noise or imbalanced given a limited number of meta-training tasks. To prevent the meta-model from being corrupted by such detrimental tasks or dominated by tasks in the majority, in this paper, we propose an adaptive task scheduler (ATS) for the meta-training process. In ATS, for the first time, we design a neural scheduler to decide which meta-training tasks to use next by predicting the probability being sampled for each candidate task, and train the scheduler to optimize the generalization capacity of the meta-model to unseen tasks. We identify two meta-model-related factors as the input of the neural scheduler, which characterize the difficulty of a candidate task to the meta-model. Theoretically, we show that a scheduler taking the two factors into account improves the meta-training loss and also the optimization landscape. Under the setting of meta-learning with noise and limited budgets, ATS improves the performance on both miniImageNet and a real-world drug discovery benchmark by up to 13% and 18%, respectively, compared to state-of-the-art task schedulers.

In this paper, we consider the problem of unsupervised domain adaptation in the semantic segmentation. There are two primary issues in this field, i.e., what and how to transfer domain knowledge across two domains. Existing methods mainly focus on adapting domain-invariant features (what to transfer) through adversarial learning (how to transfer). Context dependency is essential for semantic segmentation, however, its transferability is still not well understood. Furthermore, how to transfer contextual information across two domains remains unexplored. Motivated by this, we propose a cross-attention mechanism based on self-attention to capture context dependencies between two domains and adapt transferable context. To achieve this goal, we design two cross-domain attention modules to adapt context dependencies from both spatial and channel views. Specifically, the spatial attention module captures local feature dependencies between each position in the source and target image. The channel attention module models semantic dependencies between each pair of cross-domain channel maps. To adapt context dependencies, we further selectively aggregate the context information from two domains. The superiority of our method over existing state-of-the-art methods is empirically proved on "GTA5 to Cityscapes" and "SYNTHIA to Cityscapes".

Online Active Learning (OAL) aims to manage unlabeled datastream by selectively querying the label of data. OAL is applicable to many real-world problems, such as anomaly detection in health-care and finance. In these problems, there are two key challenges: the query budget is often limited; the ratio between classes is highly imbalanced. In practice, it is quite difficult to handle imbalanced unlabeled datastream when only a limited budget of labels can be queried for training. To solve this, previous OAL studies adopt either asymmetric losses or queries (an isolated asymmetric strategy) to tackle the imbalance, and use first-order methods to optimize the cost-sensitive measure. However, the isolated strategy limits their performance in class imbalance, while first-order methods restrict their optimization performance. In this article, we propose a novel Online Adaptive Asymmetric Active learning algorithm, based on a new asymmetric strategy (merging both asymmetric losses and queries strategies), and second-order optimization. We theoretically analyze its mistake bound and cost-sensitive metric bounds. Moreover, to better balance performance and efficiency, we enhance our algorithm via a sketching technique, which significantly accelerates the computational speed with quite slight performance degradation. Promising results demonstrate the effectiveness and efficiency of the proposed methods.

Social media has been developing rapidly in public due to its nature of spreading new information, which leads to rumors being circulated. Meanwhile, detecting rumors from such massive information in social media is becoming an arduous challenge. Therefore, some deep learning methods are applied to discover rumors through the way they spread, such as Recursive Neural Network (RvNN) and so on. However, these deep learning methods only take into account the patterns of deep propagation but ignore the structures of wide dispersion in rumor detection. Actually, propagation and dispersion are two crucial characteristics of rumors. In this paper, we propose a novel bi-directional graph model, named Bi-Directional Graph Convolutional Networks (Bi-GCN), to explore both characteristics by operating on both top-down and bottom-up propagation of rumors. It leverages a GCN with a top-down directed graph of rumor spreading to learn the patterns of rumor propagation; and a GCN with an opposite directed graph of rumor diffusion to capture the structures of rumor dispersion. Moreover, the information from source post is involved in each layer of GCN to enhance the influences from the roots of rumors. Encouraging empirical results on several benchmarks confirm the superiority of the proposed method over the state-of-the-art approaches.

We study the reinforcement learning problem of complex action control in the Multi-player Online Battle Arena (MOBA) 1v1 games. This problem involves far more complicated state and action spaces than those of traditional 1v1 games, such as Go and Atari series, which makes it very difficult to search any policies with human-level performance. In this paper, we present a deep reinforcement learning framework to tackle this problem from the perspectives of both system and algorithm. Our system is of low coupling and high scalability, which enables efficient explorations at large scale. Our algorithm includes several novel strategies, including control dependency decoupling, action mask, target attention, and dual-clip PPO, with which our proposed actor-critic network can be effectively trained in our system. Tested on the MOBA game Honor of Kings, the trained AI agents can defeat top professional human players in full 1v1 games.

Neural architecture search (NAS) has become an important approach to automatically find effective architectures. To cover all possible good architectures, we need to search in an extremely large search space with billions of candidate architectures. More critically, given a large search space, we may face a very challenging issue of space explosion. However, due to the limitation of computational resources, we can only sample a very small proportion of the architectures, which provides insufficient information for the training. As a result, existing methods may often produce sub-optimal architectures. To alleviate this issue, we propose a curriculum search method that starts from a small search space and gradually incorporates the learned knowledge to guide the search in a large space. With the proposed search strategy, our Curriculum Neural Architecture Search (CNAS) method significantly improves the search efficiency and finds better architectures than existing NAS methods. Extensive experiments on CIFAR-10 and ImageNet demonstrate the effectiveness of the proposed method.

Point-of-interest (POI) recommendation has become an increasingly important sub-field of recommendation system research. Previous methods employ various assumptions to exploit the contextual information for improving the recommendation accuracy. The common property among them is that similar users are more likely to visit similar POIs and similar POIs would like to be visited by the same user. However, none of existing methods utilize similarity explicitly to make recommendations. In this paper, we propose a new framework for POI recommendation, which explicitly utilizes similarity with contextual information. Specifically, we categorize the context information into two groups, i.e., global and local context, and develop different regularization terms to incorporate them for recommendation. A graph Laplacian regularization term is utilized to exploit the global context information. Moreover, we cluster users into different groups, and let the objective function constrain the users in the same group to have similar predicted POI ratings. An alternating optimization method is developed to optimize our model and get the final rating matrix. The results in our experiments show that our algorithm outperforms all the state-of-the-art methods.

Portfolio selection is an important yet challenging task in AI for FinTech. One of the key issues is how to represent the non-stationary price series of assets in a portfolio, which is important for portfolio decisions. The existing methods, however, fall short of capturing: 1) the complicated sequential patterns for asset price series and 2) the price correlations among multiple assets. In this paper, under a deep reinforcement learning paradigm for portfolio selection, we propose a novel Relation-aware Transformer (RAT) to handle these aspects. Specifically, being equipped with our newly developed attention modules, RAT is structurally innovated to capture both sequential patterns and asset correlations for portfolio selection. Based on the extracted sequential features, RAT is able to make profitable portfolio decisions regarding each asset via a newly devised leverage operation. Extensive experiments on real-world crypto-currency and stock datasets verify the state-of-the-art performance of RAT.

Many approaches have been proposed for time series forecasting, in light of its significance in wide applications including business demand prediction. However, the existing methods suffer from two key limitations. Firstly, most point prediction models only predict an exact value of each time step without flexibility, which can hardly capture the stochasticity of data. Even probabilistic prediction using the likelihood estimation suffers these problems in the same way. Besides, most of them use the auto-regressive generative mode, where ground-truth is provided during training and replaced by the network’s own one-step ahead output during inference, causing the error accumulation in inference. Thus they may fail to forecast time series for long time horizon due to the error accumulation. To solve these issues, in this paper, we propose a new time series forecasting model -- Adversarial Sparse Transformer (AST), based on Generated Adversarial Networks (GANs). Specifically, AST adopts a Sparse Transformer as the generator to learn a sparse attention map for time series forecasting, and uses a discriminator to improve the prediction performance from sequence level. Extensive experiments on several real-world datasets show the effectiveness and efficiency of our method.

Retrosynthesis is the process of recursively decomposing target molecules into available building blocks. It plays an important role in solving problems in organic synthesis planning. To automate or assist in the retrosynthesis analysis, various retrosynthesis prediction algorithms have been proposed. However, most of them are cumbersome and lack interpretability about their predictions. In this paper, we devise a novel template-free algorithm for automatic retrosynthetic expansion inspired by how chemists approach retrosynthesis prediction. Our method disassembles retrosynthesis into two steps: i) identify the potential reaction center of the target molecule through a novel graph neural network and generate intermediate synthons, and ii) generate the reactants associated with synthons via a robust reactant generation model. While outperforming the state-of-the-art baselines by a significant margin, our model also provides chemically reasonable interpretation.

Deep learning based medical image diagnosis has shown great potential in clinical medicine. However, it often suffers two major difficulties in real-world applications: 1) only limited labels are available for model training, due to expensive annotation costs over medical images; 2) labeled images may contain considerable label noise (e.g.,mislabeling labels) due to diagnostic difficulties of diseases. To address these, we seek to exploit rich labeled data from relevant domains to help the learning in the target task via Unsupervised Domain Adaptation (UDA). Unlike most UDA methods that rely on clean labeled data or assume samples are equally transferable, we innovatively propose a Collaborative Unsupervised Domain Adaptation algorithm, which conducts transferability-aware adaptation and conquers label noise in a collaborative way. We theoretically analyze the generalization performance of the proposed method, and also empirically evaluate it on both medical and general images. Promising experimental results demonstrate the superiority and generalization of the proposed method.

The amount of data in our society has been exploding in the era of big data. This article aims to address several open challenges in big data stream classification. Many existing studies in data mining literature follow the batch learning setting, which suffers from low efficiency and poor scalability. To tackle these challenges, we investigate a unified online learning framework for the big data stream classification task. Different from the existing online data stream classification techniques, we propose a unified Sparse Online Classification (SOC) framework. Based on SOC, we derive a second-order online learning algorithm and a cost-sensitive sparse online learning algorithm, which could successfully handle online anomaly detection tasks with the extremely unbalanced class distribution. As the performance evaluation, we analyze the theoretical bounds of the proposed algorithms and conduct an extensive set of experiments. The encouraging experimental results demonstrate the efficacy of the proposed algorithms over the state-of-the-art techniques on multiple data stream classification tasks.

Factorization Machines (FMs) learn weights for feature interactions, and achieve great success in many data mining tasks. Recently, Locally Linear Factorization Machines (LLFMs) have been proposed to capture the underlying structures of data for better performance. However, one obvious drawback of LLFM is that the local coding is only operated in the original feature space, which limits the model to be applied to high-dimensional and sparse data. In this work, we present a generalized LLFM (GLLFM) which overcomes this limitation by modeling the local coding procedure in a latent space. Moreover, a novel Supervised Variational Encoding (SVE) technique is proposed such that the distance can effectively describe the similarity between data points. Specifically, the proposed GLLFM-SVE trains several local FMs in the original space to model the higher order feature interactions effectively, where each FM associates to an anchor point in the latent space induced by SVE. The prediction for a data point is computed by a weighted sum of several local FMs, where the weights are determined by local coding coordinates with anchor points. Actually, GLLFM-SVE is quite flexible and other Neural Network (NN) based FMs can be easily embedded into this framework. Experimental results show that GLLFM-SVE significantly improves the performance of LLFM. By using NN-based FMs as local predictors, our model outperforms all the state-of-the-art methods on large-scale real-world benchmarks with similar number of parameters and comparable training time.

Traditional online multitask learning only utilizes the firstorder information of the datastream. To remedy this issue, we propose a confidence weighted multitask learning algorithm, which maintains a Gaussian distribution over each task model to guide online learning process. The mean (covariance) of the Gaussian Distribution is a sum of a local component and a global component that is shared among all the tasks. In addition, this paper also addresses the challenge of active learning on the online multitask setting. Instead of requiring labels of all the instances, the proposed algorithm determines whether the learner should acquire a label by considering the confidence from its related tasks over label prediction. Theoretical results show the regret bounds can be significantly reduced. Empirical results demonstrate that the proposed algorithm is able to achieve promising learning efficacy, while simultaneously minimizing the labeling cost.

For constrained nonconvex minimization problems, we propose a meta stochastic projection-free optimization algorithm, named Normalized Frank Wolfe Updating, that can take any Gradient Estimator (GE) as input. For this algorithm, we prove its convergence rate, regardless of the choice of GE. Using a sophisticated GE, this algorithm can significantly improve the Stochastic First order Oracle (SFO) complexity. Further, a new second order GE strategy is proposed to incorporate curvature information, which enjoys theoretical advantage over the first order ones. Besides, this paper also provides a lower bound of Linear-optimization Oracle (LO) queried to achieve an approximate stationary point. Simulation studies validate our analysis under various parameter settings.

Fog computing extends the cloud computing paradigm by allocating substantial portions of computations and services towards the edge of a network, and is, therefore, particularly suitable for large-scale, geo-distributed, and data-intensive applications. As the popularity of fog applications increases, there is a demand for the development of smart data analytic tools, which can process massive data streams in an efficient manner. To satisfy such requirements, we propose a system in which data streams generated from distributed sources are digested almost locally, whereas a relatively small amount of distilled information is converged to a center. The center extracts knowledge from the collected information, and shares it across all subordinates to boost their performances. Upon the proposed system, we devise a distributed machine learning algorithm using the online learning approach, which is well known for its high efficiency and innate ability to cope with streaming data. An asynchronous update strategy with rigorous theoretical support is applied to enhance the system robustness. Experimental results demonstrate that the proposed method is comparable with a model trained over a centralized platform in terms of the classification accuracy, whereas the efficiency and scalability of the overall system are improved.

Most state-of-the-art action localization systems process each action proposal individually, without explicitly exploiting their relations during learning. However, the relations between proposals actually play an important role in action localization, since a meaningful action always consists of multiple proposals in a video. In this paper, we propose to exploit the proposal-proposal relations using GraphConvolutional Networks (GCNs). First, we construct an action proposal graph, where each proposal is represented as a node and their relations between two proposals as an edge. Here, we use two types of relations, one for capturing the context information for each proposal and the other one for characterizing the correlations between distinct actions. Then we apply the GCNs over the graph to model the relations among different proposals and learn powerful representations for the action classification and localization. Experimental results show that our approach significantly outperforms the state-of-the-art on THUMOS14(49.1% versus 42.8%). Moreover, augmentation experiments on ActivityNet also verify the efficacy of modeling action proposal relationships.

The task of point of interest (POI) recommendation aims to recommend unvisited places to users based on their check-in history. A major challenge in POI recommendation is data sparsity, because a user typically visits only a very small number of POIs among all available POIs. In this paper, we propose AUC-MF to address the POI recommendation problem by maximizing Area Under the ROC curve (AUC). AUC has been widely used for measuring classification performance with imbalanced data distributions. To optimize AUC, we transform the recommendation task to a classification problem, where the visited locations are positive examples and the unvisited are negative ones. We define a new lambda for AUC to utilize the LambdaMF model, which combines the lambda-based method and matrix factorization model in collaborative filtering. Experiments on two datasets show that the proposed AUC-MF outperforms state-of-the-art methods significantly in terms of recommendation accuracy.

Classical semi-supervised video anomaly detection assumes that only normal data are available in the training set because of the rare and unbounded nature of anomalies. It is obviously, however, these infrequently observed abnormal events can actually help with the detection of identical or similar abnormal events, a line of thinking that motivates us to study open-set supervised anomaly detection with only a few types of abnormal observed events and many normal events available. Under the assumption that normal events can be well predicted, we propose a Margin Learning Embedded Prediction (MLEP) framework. There are three features in MLEP- based open-set supervised video anomaly detection: i) we customize a video prediction framework that favors the prediction of normal events and distorts the prediction of abnormal events; ii) The margin learning framework learns a more compact normal data distribution and enlarges the margin between normal and abnormal events. Since abnormal events are unbounded, our framework consequently helps with the detection of abnormal events, even for anomalies that have never been previously observed. Therefore, our framework is suitable for the open-set supervised anomaly detection setting; iii) our framework can readily handle both frame-level and video-level anomaly annotations. Considering that video-level anomaly detection is more easily annotated in practice and that anomaly detection with a few anomalies is a more practical setting, our work thus pushes the application of anomaly detection towards real scenarios. Extensive experiments validate the effectiveness of our framework for anomaly detection.

In this paper, we propose a decentralized distributed algorithm with stochastic communication among nodes, building on a sampling method called "edge sampling''. Such a sampling algorithm allows us to avoid the heavy peer-to-peer communication cost when combining neighboring weights on dense networks while still maintains a comparable convergence rate. In particular, we quantitatively analyze its theoretical convergence properties, as well as the optimal sampling rate over the underlying network. When compared with previous methods, our solution is shown to be unbiased, communication-efficient and suffers from lower sampling variances. These theoretical findings are validated by both numerical experiments on the mixing rates of Markov Chains and distributed machine learning problems.

Discovering disease-gene association is a fundamental and critical biomedical task, which assists biologists and physicians to discover pathogenic mechanism of syndromes. With various clinical biomarkers measuring the similarities among genes and disease phenotypes, network-based semi-supervised learning (NSSL) has been commonly utilized by these studies to address this class-imbalanced large-scale data issue. However, most existing NSSL approaches are based on linear models and suffer from two major limitations: 1) They implicitly consider a local-structure representation for each candidate; 2) They are unable to capture nonlinear associations between diseases and genes. In this paper, we propose a new framework for disease-gene association task by combining Graph Convolutional Network (GCN) and matrix factorization, named GCN-MF. With the help of GCN, we could capture non-linear interactions and exploit measured similarities. Moreover, we define a margin control loss function to reduce the effect of sparsity. Empirical results demonstrate that the proposed deep learning algorithm outperforms all other state-of-the-art methods on most of metrics.

Designing effective architectures is one of the key factors behind the success of deep neural networks. Existing deep architectures are either manually designed or automatically searched by some Neural Architecture Search (NAS) methods. However, even a well-searched architecture may still contain many non-significant or redundant modules or operations (e.g., convolution or pooling), which may not only incur substantial memory consumption and computation cost but also deteriorate the performance. Thus, it is necessary to optimize the operations inside an architecture to improve the performance without introducing extra computation cost. Unfortunately, such a constrained optimization problem is NP-hard. To make the problem feasible, we cast the optimization problem into a Markov decision process (MDP) and seek to learn a Neural Architecture Transformer (NAT) to replace the redundant operations with the more computationally efficient ones (e.g., skip connection or directly removing the connection). Based on MDP, we learn NAT by exploiting reinforcement learning to obtain the optimization policies w.r.t. different architectures. To verify the effectiveness of the proposed strategies, we apply NAT on both hand-crafted architectures and NAS based architectures. Extensive experiments on two benchmark datasets, i.e., CIFAR-10 and ImageNet, demonstrate that the transformed architecture by NAT significantly outperforms both its original form and those architectures optimized by existing methods.

Bayesian optimisation is a popular technique for hyperparameter learning but typically requires initial exploration even in cases where similar prior tasks have been solved. We propose to transfer information across tasks using learnt representations of training datasets used in those tasks. This results in a joint Gaussian process model on hyperparameters and data representations. Representations make use of the framework of distribution embeddings into reproducing kernel Hilbert spaces. The developed method has a faster convergence compared to existing baselines, in some cases requiring only a few evaluations of the target objective.

Factorization Machine (FM) is a general supervised learning framework for many AI applications due to its powerful capability of feature engineering. Despite being extensively studied, existing FM methods have several limitations in common. First of all, most existing FM methods often adopt the squared loss in the modeling process, which can be very sensitive when the data for learning contains noises and outliers. Second, some recent FM variants often explore the low-rank structure of the feature interactions matrix by relaxing the low-rank minimization problem as a trace norm minimization, which cannot always achieve a tight approximation to the original one. To address the aforementioned issues, this paper proposes a new scheme of Robust Factorization Machine (RFM) by exploring a doubly capped norms minimization approach, which employs both a capped squared trace norm in achieving a tighter approximation of the rank minimization and a capped ℓ1-norm loss to enhance the robustness of the empirical loss minimization from noisy data. We develop an efficient algorithm with a rigorous convergence proof of RFM. Experiments on public real-world datasets show that our method outperforms the state-of-the-art FM methods significantly.

Cost-Sensitive Online Classification has drawn extensive attention in recent years, where the main approach is to directly online optimize two well-known cost-sensitive metrics: (i) weighted sum of sensitivity and specificity and (ii) weighted misclassification cost. However, previous existing methods only considered first-order information of data stream. It is insufficient in practice, since many recent studies have proved that incorporating second-order information enhances the prediction performance of classification models. Thus, we propose a family of cost-sensitive online classification algorithms with adaptive regularization in this paper. We theoretically analyze the proposed algorithms and empirically validate their effectiveness and properties in extensive experiments. Then, for better trade off between the performance and efficiency, we further introduce the sketching technique into our algorithms, which significantly accelerates the computational speed with quite slight performance loss. Finally, we apply our algorithms to tackle several online anomaly detection tasks from real world. Promising results prove that the proposed algorithms are effective and efficient in solving cost-sensitive online classification problems in various real-world domains.

Points of interest (POI) recommendation has been drawn much attention recently due to the increasing popularity of location-based networks, e.g., Foursquare and Yelp. Among the existing approaches to POI recommendation, Matrix Factorization (MF) based techniques have proven to be effective. However, existing MF approaches suffer from two major problems: (1) Expensive computations and storages due to the centralized model training mechanism: the centralized learners have to maintain the whole user-item rating matrix, and potentially huge low rank matrices. (2) Privacy issues: the users' preferences are at risk of leaking to malicious attackers via the centralized learner. To solve these, we present a Decentralized MF (DMF) framework for POI recommendation. Specifically, instead of maintaining all the low rank matrices and sensitive rating data for training, we propose a random walk based decentralized training technique to train MF models on each user's end, e.g., cell phone and Pad. By doing so, the ratings of each user are still kept on one's own hand, and moreover, decentralized learning can be taken as distributed learning with multi-learners (users), and thus alleviates the computation and storage issue. Experimental results on two real-world datasets demonstrate that, comparing with the classic and state-of-the-art latent factor models, DMF significantly improvements the recommendation performance in terms of precision and recall.

Locally Linear Support Vector Machine (LLSVM) has been actively used in classification tasks due to its capability of classifying nonlinear patterns. However, existing LLSVM suffers from two drawbacks: (1) a particular and appropriate regularization for LLSVM has not yet been addressed; (2) it usually adopts a three-stage learning scheme composed of learning anchor points by clustering, learning local coding coordinates by a predefined coding scheme, and finally learning for training classifiers. We argue that this decoupled approaches oversimplifies the original optimization problem, resulting in a large deviation due to the disparate purpose of each step. To address the first issue, we propose a novel diversified regularization which could capture infrequent patterns and reduce the model size without sacrificing the representation power. Based on this regularization, we develop a joint optimization algorithm among anchor points, local coding coordinates and classifiers to simultaneously minimize the overall classification risk, which is termed as Diversified and Unified Locally Linear Support Vector Machine (DU-LLSVM for short). To the best of our knowledge, DU-LLSVM is the first principled method that directly learns sparse local coding and can be easily generalized to other supervised learning models. Extensive experiments showed that DU-LLSVM consistently surpassed several state-of-the-art methods with a predefined local coding scheme (e.g. LLSVM) or a supervised anchor point learning (e.g. SAPL-LLSVM).

Recently, the decentralized optimization problem is attracting growing attention. Most existing methods are deterministic with high per-iteration cost and have a convergence rate quadratically depending on the problem condition number. Besides, the dense communication is necessary to ensure the convergence even if the dataset is sparse. In this paper, we generalize the decentralized optimization problem to a monotone operator root finding problem, and propose a stochastic algorithm named DSBA that (1) converges geometrically with a rate linearly depending on the problem condition number, and (2) can be implemented using sparse communication only. Additionally, DSBA handles important learning problems like AUC-maximization which can not be tackled efficiently in the previous problem setting. Experiments on convex minimization and AUC-maximization validate the efficiency of our method.

Traditional online learning on graphs adapts graph Laplacian into ridge regression, which may not guarantee reasonable accuracy when the data are adversarially generated. To solve this issue, we exploit an adaptive optimization framework for online classification on graphs. The derived model can achieve a min-max regret under an adversarial mechanism of data generation. To take advantage of the informative labels, we propose an adaptive large-margin update rule, which enjoys a lower regret than the algorithms using error-driven update rules. However, this algorithm assumes that the full information label is provided for each node, which is violated in many practical applications where labeling is expensive and the oracle may only tell whether the prediction is correct or not. To address this issue, we propose a bandit online algorithm on graphs. It derives per-instance confidence region of the prediction, from which the model can be learned adaptively to minimize the online regret. Experiments on benchmark graph datasets show that the proposed bandit algorithm outperforms state-of-the-art competitors, even sometimes beats the algorithms using full information label feedback.

We investigate how to adopt dual random projection for high-dimensional similarity learning. For a high-dimensional similarity learning problem, projection is usually adopted to map high-dimensional features into low-dimensional space, in order to reduce the computational cost. However, dimensionality reduction method sometimes results in unstable performance due to the suboptimal solution in original space. In this paper, we propose a dual random projection framework for similarity learning to recover the original optimal solution from subspace optimal solution. Previous dual random projection methods usually make strong assumptions about the data, which need to be low rank or have a large margin. Those assumptions limit dual random projection applications in similarity learning. Thus, we adopt a dual-sparse regularized random projection method that introduces a sparse regularizer into the reduced dual problem. As the original dual solution is a sparse one, applying a sparse regularizer in the reduced space relaxes the low-rank assumption. Experimental results show that our method enjoys higher effectiveness and efficiency than state-of-the-art solutions.

Collaborative filtering, especially latent factor model, has been popularly used in personalized recommendation. Latent factor model aims to learn user and item latent factors from user-item historic behaviors. To apply it into real big data scenarios, efficiency becomes the first concern, including offline model training efficiency and online recommendation efficiency. In this paper, we propose a D istributed C ollaborative H ashing ( DCH ) model which can significantly improve both efficiencies. Specifically, we first propose a distributed learning framework, following the state-of-the-art parameter server paradigm, to learn the offline collaborative model. Our model can be learnt efficiently by distributedly computing subgradients in minibatches on workers and updating model parameters on servers asynchronously. We then adopt hashing technique to speedup the online recommendation procedure. Recommendation can be quickly made through exploiting lookup hash tables. We conduct thorough experiments on two real large-scale datasets. The experimental results demonstrate that, comparing with the classic and state-of-the-art (distributed) latent factor models, DCH has comparable performance in terms of recommendation accuracy but has both fast convergence speed in offline model training procedure and realtime efficiency in online recommendation procedure. Furthermore, the encouraging performance of DCH is also shown for several real-world applications in Ant Financial.

This paper investigates Online Active Learning (OAL) for imbalanced unlabeled datastream, where only a budget of labels can be queried to optimize some cost-sensitive performance measure. OAL can solve many real-world problems, such as anomaly detection in healthcare, finance and network security. In these problems, there are two key challenges: the query budget is often limited; the ratio between two classes is highly imbalanced. To address these challenges, existing work of OAL adopts either asymmetric losses or queries (an isolated asymmetric strategy) to tackle the imbalance, and uses first-order methods to optimize the cost-sensitive measure. However, they may incur two deficiencies: (1) the poor ability in handling imbalanced data due to the isolated asymmetric strategy; (2) relative slow convergence rate due to the first-order optimization. In this paper, we propose a novel Online Adaptive Asymmetric Active (OA3) learning algorithm, which is based on a new asymmetric strategy (merging both the asymmetric losses and queries strategies), and second-order optimization. We theoretically analyze its bounds, and also empirically evaluate it on four real-world online anomaly detection tasks. Promising results confirm the effectiveness and robustness of the proposed algorithm in various application domains.

This paper aims at improvement on the effectiveness of matrix decomposition (MD) methods for implicit feedback. We highlight two critical limitations of existing works. First, due to the large number of unlabeled feedback, most existing works employ a uniform weight to the missing data to reduce computational complexity. However, such a uniform assumption may rarely hold in real-world scenarios. Second, the commonly-used bilateral loss function might be infinite if the data point is mis-classified. Outliers may have such issues and misguide the learning process. We address the above two issues by learning a robust asymmetric learning model. By leveraging the cost-sensitive learning and capped unilateral loss function, our robust MD objective function integrates them into a joint formulation, where the low-rank basis for user/item profiles can be modeled in an effective and robust way. Particularly, a novel log-determinant function is employed to refine the nuclear norm with respect to the low-rank approximation. We derive an iterative re-weighted algorithm to efficiently minimize this MD objective, and also rigorously prove a lower error bound of the proposed algorithm compared to the 1-bit matrix completion method. Finally, we show the promising experimental results of our algorithm on benchmark recommendation datasets.

Recommender systems with implicit feedback (e.g. clicks and purchases) suffer from two critical limitations: 1) imbalanced labels may mislead the learning process of the conventional models that assign balanced weights to the classes; and 2) outliers with large reconstruction errors may dominate the objective function by the conventional $L_2$-norm loss. To address these issues, we propose a robust asymmetric recommendation model. It integrates cost-sensitive learning with capped unilateral loss into a joint objective function, which can be optimized by an iteratively weighted approach. To reduce the computational cost of low-rank approximation, we exploit the dual characterization of the nuclear norm to derive a min-max optimization problem and design a subgradient algorithm without performing full SVD. Finally, promising empirical results demonstrate the effectiveness of our algorithm on benchmark recommendation datasets.

Factorization Machine (FM) is a supervised learning approach with a powerful capability of feature engineering. It yields state-of-the-art performances in various batch learning tasks where all the training data is made available prior to the training. However, in real-world applications where the data arrives sequentially in a streaming manner, the high cost of re-training with batch learning algorithms has posed formidable challenges in the online learning scenario. The initial challenge is that no prior formulations of FM could directly fulfill the requirements in Online Convex Optimization (OCO) -- the paramount framework for online learning algorithm design. To address this aforementioned challenge, we invent a new convexification scheme leading to a Compact Convexified FM (CCFM) that seamlessly meets the requirements in OCO. However for learning Compact Convexified FM (CCFM) in the online learning settings, most existing algorithms suffer from expensive projection operations. To address this subsequent challenge, we follow the general projection-free algorithmic framework of Online Conditional Gradient and propose an Online Compact Convex Factorization Machine (OCCFM) algorithm that eschews the projection operation with efficient linear optimization steps. In support of the proposed OCCFM in terms of its theoretical foundation, we prove that the developed algorithm achieves a sub-linear regret bound. To evaluate the empirical performance of OCCFM, we conduct extensive experiments on 6 real-world datasets for online regression and online classification tasks. The experimental results show that OCCFM outperforms the state-of-art online learning methods for FM.

One critical deficiency of traditional online kernel learning methods is their unbounded and growing number of support vectors in the online learning process, making them inefficient and non-scalable for large-scale applications. Recent studies on scalable online kernel learning have attempted to overcome this shortcoming, e.g., by imposing a constant budget on the number of support vectors. Although they attempt to bound the number of support vectors at each online learning iteration, most of them fail to bound the number of support vectors for the final output hypothesis, which is often obtained by averaging the series of hypotheses over all the iterations. In this article, we propose a novel framework for bounded online kernel methods, named “Sparse Passive-Aggressive (SPA)” learning, which is able to yield a final output kernel-based hypothesis with a bounded number of support vectors. Unlike the common budget maintenance strategy used by many existing budget online kernel learning approaches, the idea of our approach is to attain the bounded number of support vectors using an efficient stochastic sampling strategy that samples an incoming training example as a new support vector with a probability proportional to its loss suffered. We theoretically prove that SPA achieves an optimal mistake bound in expectation, and we empirically show that it outperforms various budget online kernel learning algorithms. Finally, in addition to general online kernel learning tasks, we also apply SPA to derive bounded online multiple-kernel learning algorithms, which can significantly improve the scalability of traditional Online Multiple-Kernel Classification (OMKC) algorithms while achieving satisfactory learning accuracy as compared with the existing unbounded OMKC algorithms.

In literature, learning with expert advice methods usually assume that a learner always obtain the true label of every incoming training instance at the end of each trial. However, in many real-world applications, acquiring the true labels of all instances can be both costly and time consuming, especially for large-scale problems. For example, in the social media, data stream usually comes in a high speed and volume, and it is nearly impossible and highly costly to label all of the instances. In this article, we address this problem with active learning with expert advice, where the ground truth of an instance is disclosed only when it is requested by the proposed active query strategies. Our goal is to minimize the number of requests while training an online learning model without sacrificing the performance. To address this challenge, we propose a framework of active forecasters, which attempts to extend two fully supervised forecasters, Exponentially Weighted Average Forecaster and Greedy Forecaster, to tackle the task of online active learning (OAL) with expert advice. Specifically, we proposed two OAL with expert advice algorithms, named Active Exponentially Weighted Average Forecaster (AEWAF) and active greedy forecaster (AGF), by considering the difference of expert advices. To further improve the robustness of the proposed AEWAF and AGF algorithms in the noisy scenarios (where noisy experts exist), we also proposed two robust active learning with expert advice algorithms, named Robust Active Exponentially Weighted Average Forecaster and Robust Active Greedy Forecaster. We validate the efficacy of the proposed algorithms by an extensive set of experiments in both normal scenarios (where all of experts are comparably reliable) and noisy scenarios.

The goal of online active learning is to learn predictive models from a sequence of unlabeled data given limited label query budget. Unlike conventional online learning tasks, online active learning is considerably more challenging because of two reasons. First, it is difficult to design an effective query strategy to decide when is appropriate to query the label of an incoming instance given limited query budget. Second, it is also challenging to decide how to update the predictive models effectively whenever the true label of an instance is queried. Most existing approaches for online active learning are often based on a family of first-order online learning algorithms, which are simple and efficient but fall short in the slow convergence and sub-optimal solution in exploiting the labeled training data. To solve these issues, this paper presents a novel framework of Second-order Online Active Learning (SOAL) by fully exploiting both the first-order and second-order information. The proposed algorithms are able to achieve effective online learning efficacy, maximize the predictive accuracy, and minimize the labeling cost. To make SOAL more practical for real-world applications, especially for class-imbalanced online classification tasks (e.g., malicious web detection), we extend the SOAL framework by proposing the Costsensitive Second-order Online Active Learning algorithm named “SOALCS”, which is devised by maximizing the sum of weighted sensitivity and specificity or minimizing the cost of weighted mistakes of different classes. We conducted both theoretical analysis and empirical studies, including an extensive set of experiments on a variety of large-scale real-world datasets, in which the promising empirical results validate the efficacy and scalability of the proposed algorithms towards large-scale online learning tasks.

Learning to optimize AUC performance for classifying label imbalanced data in online scenarios has been extensively studied in recent years. Most of the existing work has attempted to address the problem directly in the original feature space, which may not suitable for non-linearly separable datasets. To solve this issue, some kernel-based learning methods are proposed for non-linearly separable datasets. However, such kernel approaches have been shown to be inefficient and failed to scale well on large scale datasets in practice. Taking this cue, in this work, we explore the use of scalable kernel-based learning techniques as surrogates to existing approaches: random Fourier features and Nyström method, for tackling the problem and bring insights to the differences between the two methods based on their online performance. In contrast to the conventional kernel-based learning methods which suffer from high computational complexity of the kernel matrix, our proposed approaches elevate this issue with linear features that approximate the kernel function/matrix. Specifically, two different surrogate kernel-based learning models are presented for addressing the online AUC maximization task: (i) the Fourier Online AUC Maximization (FOAM) algorithm that samples the basis functions from a data-independent distribution to approximate the kernel functions; and (ii) the Nyström Online AUC Maximization (NOAM) algorithm that samples a subset of instances from the training data to approximate the kernel matrix by a low rank matrix. Another novelty of the present work is the proposed mini-batch Online Gradient Descent method for model updating to control the noise and reduce the variance of gradients. We provide theoretical analyses for the two proposed algorithms. Empirical studies on commonly used large scale datasets show that the proposed algorithms outperformed existing state-of-the-art methods in terms of both AUC performance and computational efficiency.

Precipitation prediction, such as short-term rainfall prediction, is a very important problem in the field of meteorological service. In practice, most of recent studies focus on leveraging radar data or satellite images to make predictions. However, there is another scenario where a set of weather features are collected by various sensors at multiple observation sites. The observations of a site are sometimes incomplete but provide important clues for weather prediction at nearby sites, which are not fully exploited in existing work yet. To solve this problem, we propose a multi-task convolutional neural network model to automatically extract features from the time series measured at observation sites and leverage the correlation between the multiple sites for weather prediction via multi-tasking. To the best of our knowledge, this is the first attempt to use multi-task learning and deep learning techniques to predict short-term rainfall amount based on multi-site features. Specifically, we formulate the learning task as an end-to-end multi-site neural network model which allows to leverage the learned knowledge from one site to other correlated sites, and model the correlations between different sites. Extensive experiments show that the learned site correlations are insightful and the proposed model significantly outperforms a broad set of baseline models including the European Centre for Medium-range Weather Forecasts system (ECMWF).

The conditional gradient algorithm has regained a surge of research interest in recent years due to its high efficiency in handling large-scale machine learning problems. However, none of existing studies has explored it in the distributed online learning setting, where locally light computation is assumed. In this paper, we fill this gap by proposing the distributed online conditional gradient algorithm, which eschews the expensive projection operation needed in its counterpart algorithms by exploiting much simpler linear optimization steps. We give a regret bound for the proposed algorithm as a function of the network size and topology, which will be smaller on smaller graphs or “well-connected” graphs. Experiments on two large-scale real-world datasets for a multiclass classification task confirm the computational benefit of the proposed algorithm and also verify the theoretical regret bound.

Relative similarity learning~(RSL) aims to learn similarity functions from data with relative constraints. Most previous algorithms developed for RSL are batch-based learning approaches which suffer from poor scalability when dealing with real-world data arriving sequentially. These methods are often designed to learn a single similarity function for a specific task. Therefore, they may be sub-optimal to solve multiple task learning problems. To overcome these limitations, we propose a scalable RSL framework named OMTRSL (Online Multi-Task Relative Similarity Learning). Specifically, we first develop a simple yet effective online learning algorithm for multi-task relative similarity learning. Then, we also propose an active learning algorithm to save the labeling cost. The proposed algorithms not only enjoy theoretical guarantee, but also show high efficacy and efficiency in extensive experiments on real-world datasets.

Social recommender systems exploit users' social relationships to improve recommendation accuracy. Intuitively, a user tends to trust different people regarding with different scenarios. Therefore, one main challenge of social recommendation is to exploit the most appropriate dependencies between users for a given recommendation task. Previous social recommendation methods are usually developed based on pre-defined user dependencies. Thus, they may not be optimal for a specific recommendation task. In this paper, we propose a novel recommendation method, named probabilistic relational matrix factorization (PRMF), which can automatically learn the dependencies between users to improve recommendation accuracy. In PRMF, users' latent features are assumed to follow a matrix variate normal (MVN) distribution. Both positive and negative user dependencies can be modeled by the row precision matrix of the MVN distribution. Moreover, we also propose an alternating optimization algorithm to solve the optimization problem of PRMF. Extensive experiments on four real datasets have been performed to demonstrate the effectiveness of the proposed PRMF model.

Factorization Machines (FMs) are a widely used method for efficiently using high-order feature interactions in classification and regression tasks. Unfortunately, despite increasing interests in FMs, existing work only considers high order information of the input features which limits their capacities in non-linear problems and fails to capture the underlying structures of more complex data. In this work, we present a novel Locally Linear Factorization Machines (LLFM) which overcomes this limitation by exploring local coding technique. Unlike existing local coding classifiers that involve a phase of unsupervised anchor point learning and predefined local coding scheme which is suboptimal as the class label information is not exploited in discovering the encoding and thus can result in a suboptimal encoding for prediction, we formulate a joint optimization over the anchor points, local coding coordinates and FMs variables to minimize classification or regression risk. Empirically, we demonstrate that our approach achieves much better predictive accuracy than other competitive methods which employ LLFM with unsupervised anchor point learning and predefined local coding scheme.

In recent years, due to the emergence of Big Data (terabytes or petabytes) and Big Model (tens of billions of parameters), there has been an ever-increasing need of parallelizing machine learning (ML) algorithms in both academia and industry. Although there are some existing distributed computing systems, such as Hadoop and Spark, for parallelizing ML algorithms, they only provide synchronous and coarse-grained operators (e.g., Map, Reduce, and Join, etc.), which may hinder developers from implementing more efficient algorithms. This motivated us to design a universal distributed platform termed KunPeng, that combines both distributed systems and parallel optimization algorithms to deal with the complexities that arise from large-scale ML. Specifically, KunPeng not only encapsulates the characteristics of data/model parallelism, load balancing, model sync-up, sparse representation, industrial fault-tolerance, etc., but also provides easy-to-use interface to empower users to focus on the core ML logics. Empirical results on terabytes of real datasets with billions of samples and features demonstrate that, such a design brings compelling performance improvements on ML programs ranging from Follow-the-Regularized-Leader Proximal algorithm to Sparse Logistic Regression and Multiple Additive Regression Trees. Furthermore, KunPeng's encouraging performance is also shown for several real-world applications including the Alibaba's Double 11 Online Shopping Festival and Ant Financial's transaction risk estimation.

Experimental determination of drug-target interactions is expensive and time-consuming. Therefore, there is a continuous demand for more accurate predictions of interactions using computational techniques. Algorithms have been devised to infer novel interactions on a global scale where the input to these algorithms is a drug-target network (i.e., a bipartite graph where edges connect pairs of drugs and targets that are known to interact). However, these algorithms had difficulty predicting interactions involving new drugs or targets for which there are no known interactions (i.e., "orphan" nodes in the network). Since data usually lie on or near to low-dimensional non-linear manifolds, we propose two matrix factorization methods that use graph regularization in order to learn such manifolds. In addition, considering that many of the non-occurring edges in the network are actually unknown or missing cases, we developed a preprocessing step to enhance predictions in the "new drug" and "new target" cases by adding edges with intermediate interaction likelihood scores. In our cross validation experiments, our methods achieved better results than three other state-of-the-art methods in most cases. Finally, we simulated some "new drug" and "new target" cases and found that GRMF predicted the left-out interactions reasonably well.

Multi-Task Learning (MTL) can enhance a classifier's generalization performance by learning multiple related tasks simultaneously. Conventional MTL works under the offline or batch setting, and suffers from expensive training cost and poor scalability. To address such inefficiency issues, online learning techniques have been applied to solve MTL problems. However, most existing algorithms of online MTL constrain task relatedness into a presumed structure via a single weight matrix, which is a strict restriction that does not always hold in practice. In this paper, we propose a robust online MTL framework that overcomes this restriction by decomposing the weight matrix into two components: The first one captures the low-rank common structure among tasks via a nuclear norm and the second one identifies the personalized patterns of outlier tasks via a group lasso. Theoretical analysis shows the proposed algorithm can achieve a sub-linear regret with respect to the best linear model in hindsight. Even though the above framework achieves good performance, the nuclear norm that simply adds all nonzero singular values together may not be a good low-rank approximation. To improve the results, we use a log-determinant function as a non-convex rank approximation. The gradient scheme is applied to optimize log-determinant function and can obtain a closed-form solution for this refined problem. Experimental results on a number of real-world applications verify the efficacy of our method.

Autoregressive integrated moving average (ARIMA) is one of the most popular linear models for time series forecasting due to its nice statistical properties and great flexibility. However, its parameters are estimated in a batch manner and its noise terms are often assumed to be strictly bounded, which restricts its applications and makes it inefficient for handling large-scale real data. In this paper, we propose online learning algorithms for estimating ARIMA models under relaxed assumptions on the noise terms, which is suitable to a wider range of applications and enjoys high computational efficiency. The idea of our ARIMA method is to reformulate the ARIMA model into a task of full information online optimization (without random noise terms). As a consequence, we can online estimation of the parameters in an efficient and scalable way. Furthermore, we analyze regret bounds of the proposed algorithms, which guarantee that our online ARIMA model is provably as good as the best ARIMA model in hindsight. Finally, our encouraging experimental results further validate the effectiveness and robustness of our method.

Mining data streams has been an important open research problem in the era of big data analytics. This paper investigates supervised machine learning techniques for mining data streams with application to online anomaly detection. Unlike conventional data mining tasks, mining data streams for online anomaly detection has several challenges: (i) data arriving sequentially and increasing rapidly, (ii) highly class-imbalanced distributions; and (iii) complex anomaly patterns that could evolve dynamically. To tackle these challenges, we propose Cost-Sensitive Online Multiple Kernel Classification (CSOMKC) for comprehensively mining data streams and demonstrate its application to online anomaly detection. Specifically, CSOMKC learns a kernel-based cost-sensitive prediction model for imbalanced data streams in a sequential or online learning fashion, in which a pool of multiple diverse kernels is dynamically explored. The optimal kernel predictor and the multiple kernel combination are learnt together, and simultaneously class imbalance issues are addressed. We perform theoretical and extensive empirical analysis of the proposed algorithms.

This paper aims to investigate efficient and scalable machine learning algorithms for resolving Non-negative Matrix Factorization (NMF), which is important for many real-world applications, particularly for collaborative filtering and recommender systems. Unlike traditional batch learning methods, a recently proposed online learning technique named "NN-PA" tackles NMF by applying the popular Passive-Aggressive (PA) online learning, and found promising results. Despite its simplicity and high efficiency, NN-PA falls short in at least two critical limitations: (i) it only exploits the first-order information and thus may converge slowly especially at the beginning of online learning tasks; (ii) it is sensitive to some key parameters which are often difficult to be tuned manually, particularly in a practical online learning system. In this work, we present a novel family of online Adaptive Passive-Aggressive (APA) learning algorithms for NMF, named "NN-APA", which overcomes two critical limitations of NN-PA by (i) exploiting second-order information to enhance PA in making more informative updates at each iteration; and (ii) achieving the parameter auto-selection by exploring the idea of online learning with expert advice in deciding the optimal combination of the key parameters in NMF. We theoretically analyze the regret bounds of the proposed method and show its advantage over the state-of-the-art NN-PA method, and further validate the efficacy and scalability of the proposed technique through an extensive set of experiments on a variety of large-scale real recommender systems datasets.

This paper investigates the problem of online active learning for training classification models from sequentially arriving data. This is more challenging than conventional online learning tasks since the learner not only needs to figure out how to effectively update the classifier but also needs to decide when is the best time to query the label of an incoming instance given limited label budget. The existing online active learning approaches are often based on first-order online learning methods which generally fall short in slow convergence rate and sub-optimal exploitation of available information when querying the labeled data. To overcome the limitations, in this paper, we present a new framework of Second-order Online Active Learning (SOAL), which fully exploits both first-order and second-order information to achieve high learning accuracy with low labeling cost. We conduct both theoretical analysis and empirical studies for evaluating the proposed SOAL algorithm extensively. The encouraging results show clear advantages of the proposed algorithm over a family of state-of-the-art online active learning algorithms.

A series of online multi-task learning (OMTL) algorithms have been proposed to avoid the expensive training cost and poor adaptability of traditional batch multi-task learning (MTL) algorithms in recent years. However, these OMTL algorithms usually assume that all tasks are closely related, which may not hold in practical scenarios. More importantly, their theoretical reliability is weakened due to the lack of proof on the cumulative regrets. To overcome these limitations, we present a robust online multi-task classification framework (ROM) and its two optimization algorithms (ROM-PGD, ROM-RDA). The proposed algorithms can not only automatically capture the common features among all tasks and individual features for each task, but also identify the potential existence of outlier task. Theoretically, we prove that the regret bounds of these two algorithms are sub-linear compared with the best separating algorithm in hindsight. Empirical studies on both synthetic and real-world datasets also demonstrate the effectiveness of our proposed algorithms when compared with the state-of-the-art OMTL algorithms.

Matrix eigen-decomposition is a classic and long-standing problem that plays a fundamental role in scientific computing and machine learning. Despite some existing algorithms for this inherently non-convex problem, the study remains inadequate for the need of large data nowadays. To address this gap, we propose a Doubly Stochastic Riemannian Gradient EIGenSolver, DSRG-EIGS, where the double stochasticity comes from the generalization of the stochastic Euclidean gradient ascent and the stochastic Euclidean coordinate ascent to Riemannian manifolds. As a result, it induces a greatly reduced complexity per iteration, enables the algorithm to completely avoid the matrix inversion, and consequently makes it well-suited to large-scale applications. We theoretically analyze its convergence properties and empirically validate it on real-world datasets. Encouraging experimental results demonstrate its advantages over the deterministic counterparts.

A large number of new buyers are often acquired by merchants during promotions. However, many of the attracted buyers are one-time deal hunters, and the promotions may have little long-lasting impact on sales. It is important for merchants to identify who can be converted to regular loyal buyers and then target them to reduce promotion cost and increase the return on investment (ROI). At International Joint Conferences on Artificial Intelligence (IJCAI) 2015, Alibaba hosted an international competition for repeat buyer prediction based on the sales data of the ``Double 11" shopping event in 2014 at Tmall.com. We won the first place at stage 1 of the competition out of 753 teams. In this paper, we present our winning solution, which consists of comprehensive feature engineering and model training. We created profiles for users, merchants, brands, categories, items and their interactions via extensive feature engineering. These profiles are not only useful for this particular prediction task, but can also be used for other important tasks in e-commerce, such as customer segmentation, product recommendation, and customer base augmentation for brands. Feature engineering is often the most important factor for the success of a prediction task, but not much work can be found in the literature on feature engineering for prediction tasks in e-commerce. Our work provides some useful hints and insights for data science practitioners in e-commerce.

Conventional online kernel methods often yield an unbounded large number of support vectors, making them inefficient and non-scalable for large-scale applications. Recent studies on bounded kernel-based online learning have attempted to overcome this shortcoming. Although they can bound the number of support vectors at each iteration, most of them fail to bound the number of support vectors for the final output solution which is often obtained by averaging the series of solutions over all the iterations. In this paper, we propose a novel kernel-based online learning method, Sparse Passive Aggressive learning (SPA), which can output a final solution with a bounded number of support vectors. The key idea of our method is to explore an efficient stochastic sampling strategy, which turns an example into a new support vector with some probability that depends on the loss suffered by the example. We theoretically prove that the proposed SPA algorithm achieves an optimal regret bound in expectation, and empirically show that the new algorithm outperforms various bounded kernel-based online learning algorithms.

Detecting temporal patterns is one of the most prevalent challenges while mining data. Often, timestamps or information about when certain instances or events occurred can provide us with critical information to recognize temporal patterns. Unfortunately, most existing techniques are not able to fully extract useful temporal information based on the time (especially at different resolutions of time). They miss out on 3 crucial factors: (i) they do not distinguish between timestamp features (which have cyclical or periodic properties) and ordinary features; (ii) they are not able to detect patterns exhibited at different resolutions of time (e.g. different patterns at the annual level, and at the monthly level); and (iii) they are not able to relate different features (e.g. multimodal features) of instances with different temporal properties (e.g. while predicting stock prices, stock fundamentals may have annual patterns, and at the same time factors like peer stock prices and global markets may exhibit daily patterns). To solve these issues, we offer a novel multiple-kernel learning view and develop Temporal Kernel Descriptors which utilize Kernel functions to comprehensively detect temporal patterns by deriving relationship of instances with the time features. We automatically learn the optimal kernel function, and hence the optimal temporal similarity between two instances. We formulate the optimization as a Multiple Kernel Learning (MKL) problem. We empirically evaluate its performance by solving the optimization using Online MKL.

Multi-Task Learning (MTL) can enhance the classifier's generalization performance by learning multiple related tasks simultaneously. Conventional MTL works under the offline or batch learning setting and suffers from the expensive training cost together with the poor scalability. To address such inefficiency issues, online learning technique has been applied to solve MTL problems. However, most existing algorithms for online MTL constrain task relatedness into a presumed structure via a single weight matrix, a strict restriction that does not always hold in practice. In this paper, we propose a general online MTL framework that overcomes this restriction by decomposing the weight matrix into two components: the first component captures the correlative structure among tasks in a low-rank subspace, and the second component identifies the personalized patterns for the outlier tasks. A projected gradient scheme is devised to learn such components adaptively. Theoretical analysis shows that the proposed algorithm can achieve a sub-linear regret with respect to the best linear model in hindsight. Experimental investigation on a number of real-world datasets also verifies the efficacy of our approach.

In this paper, we present a new framework for large scale online kernel learning, making kernel methods efficient and scalable for large-scale online learning applications. Unlike the regular budget online kernel learning scheme that usually uses some budget maintenance strategies to bound the number of support vectors, our framework explores a completely different approach of kernel functional approximation techniques to make the subsequent online learning task efficient and scalable. Specifically, we present two different online kernel machine learning algorithms: (i) Fourier Online Gradient Descent (FOGD) algorithm that applies the random Fourier features for approximating kernel functions; and (ii) Nyström Online Gradient Descent (NOGD) algorithm that applies the Nyström method to approximate large kernel matrices. We explore these two approaches to tackle three online learning tasks: binary classification, multi-class classification, and regression. The encouraging results of our experiments on large-scale datasets validate the effectiveness and efficiency of the proposed algorithms, making them potentially more practical than the family of existing budget online kernel learning approaches.

In pharmaceutical sciences, a crucial step of the drug discovery process is the identification of drug-target interactions. However, only a small portion of the drug-target interactions have been experimentally validated, as the experimental validation is laborious and costly. To improve the drug discovery efficiency, there is a great need for the development of accurate computational approaches that can predict potential drug-target interactions to direct the experimental verification. In this paper, we propose a novel drug-target interaction prediction algorithm, namely neighborhood regularized logistic matrix factorization (NRLMF). Specifically, the proposed NRLMF method focuses on modeling the probability that a drug would interact with a target by logistic matrix factorization, where the properties of drugs and targets are represented by drug-specific and target-specific latent vectors, respectively. Moreover, NRLMF assigns higher importance levels to positive observations (i.e., the observed interacting drug-target pairs) than negative observations (i.e., the unknown pairs). Because the positive observations are already experimentally verified, they are usually more trustworthy. Furthermore, the local structure of the drug-target interaction data has also been exploited via neighborhood regularization to achieve better prediction accuracy. We conducted extensive experiments over four benchmark datasets, and NRLMF demonstrated its effectiveness compared with five state-of-the-art approaches.

Online learning plays an important role in many big data mining problems because of its high efficiency and scalability. In the literature, many online learning algorithms using gradient information have been applied to solve online classification problems. Recently, more effective second-order algorithms have been proposed, where the correlation between the features is utilized to improve the learning efficiency. Among them, Confidence-Weighted (CW) learning algorithms are very effective, which assume that the classification model is drawn from a Gaussian distribution, which enables the model to be effectively updated with the second-order information of the data stream. Despite being studied actively, these CW algorithms cannot handle nonseparable datasets and noisy datasets very well. In this article, we propose a family of Soft Confidence-Weighted (SCW) learning algorithms for both binary classification and multiclass classification tasks, which is the first family of online classification algorithms that enjoys four salient properties simultaneously: (1) large margin training, (2) confidence weighting, (3) capability to handle nonseparable data, and (4) adaptive margin. Our experimental results show that the proposed SCW algorithms significantly outperform the original CW algorithm. When comparing with a variety of state-of-the-art algorithms (including AROW, NAROW, and NHERD), we found that SCW in general achieves better or at least comparable predictive performance, but enjoys considerably better efficiency advantage (i.e., using a smaller number of updates and lower time cost). To facilitate future research, we release all the datasets and source code to the public at http://libol.stevenhoi.org/.

Distance metric learning (DML) is an important technique to improve similarity search in content-based image retrieval. Despite being studied extensively, most existing DML approaches typically adopt a single-modal learning framework that learns the distance metric on either a single feature type or a combined feature space where multiple types of features are simply concatenated. Such single-modal DML methods suffer from some critical limitations: (i) some type of features may significantly dominate the others in the DML task due to diverse feature representations; and (ii) learning a distance metric on the combined high-dimensional feature space can be extremely time-consuming using the naive feature concatenation approach. To address these limitations, in this paper, we investigate a novel scheme of online multi-modal distance metric learning (OMDML), which explores a unified two-level online learning scheme: (i) it learns to optimize a distance metric on each individual feature space; and (ii) then it learns to find the optimal combination of diverse types of features. To further reduce the expensive cost of DML on high-dimensional feature space, we propose a low-rank OMDML algorithm which not only significantly reduces the computational cost but also retains highly competing or even better learning accuracy. We conduct extensive experiments to evaluate the performance of the proposed algorithms for multi-modal image retrieval, in which encouraging results validate the effectiveness of the proposed technique.

Learning for maximizing AUC performance is an important research problem in machine learning. Unlike traditional batch learning methods for maximizing AUC which often suffer from poor scalability, recent years have witnessed some emerging studies that attempt to maximize AUC by single-pass online learning approaches. Despite their encouraging results reported, the existing online AUC maximization algorithms often adopt simple stochastic gradient descent approaches, which fail to exploit the geometry knowledge of the data observed in the online learning process, and thus could suffer from relatively slow convergence. To overcome the limitation of the existing studies, in this paper, we propose a novel algorithm of Adaptive Online AUC Maximization (AdaOAM), by applying an adaptive gradient method for exploiting the knowledge of historical gradients to perform more informative online learning. The new adaptive updating strategy by AdaOAM is less sensitive to parameter settings due to its natural effect of tuning the learning rate. In addition, the time complexity of the new algorithm remains the same as the previous non-adaptive algorithms. To demonstrate the effectiveness of the proposed algorithm, we analyze its theoretical bound, and further evaluate its empirical performance on both public benchmark datasets and anomaly detection datasets. The encouraging empirical results clearly show the effectiveness and efficiency of the proposed algorithm.

Relative similarity learning, as an important learning scheme for information retrieval, aims to learn a bi-linear similarity function from a collection of labeled instance-pairs, and the learned function would assign a high similarity value for a similar instance-pair and a low value for a dissimilar pair. Existing algorithms usually assume the labels of all the pairs in data streams are always made available for learning. However, this is not always realistic in practice since the number of possible pairs is quadratic to the number of instances in the database, and manually labeling the pairs could be very costly and time consuming. To overcome the limitation, we propose a novel framework of active online similarity learning. Specifically, we propose two new algorithms: (i)~PAAS: Passive-Aggressive Active Similarity learning; (ii)~CWAS: Confidence-Weighted Active Similarity learning, and we will prove their mistake bounds in theory. We have conducted extensive experiments on a variety of real-world data sets, and we find encouraging results that validate the empirical effectiveness of the proposed algorithms.

Traditional online learning for graph node classification adapts graph regularization into ridge regression, which may not be suitable when data is adversarially generated. To solve this issue, we propose a more general min-max optimization framework for online graph node classification. The derived online algorithm can achieve a min-max regret compared with the optimal linear model found offline. However, this algorithm assumes that the label is provided for every node, while label is scare and labeling is usually either too time-consuming or expensive in real-world applications. To save labeling effort, we propose a novel confidence-based query approach to prioritize the informative labels. Our theoretical result shows that an online algorithm learning on these selected labels can achieve comparable mistake bound with the fully-supervised online counterpart. To take full advantage of these labels, we propose an aggressive algorithm, which can update the model even if no error occurs. Theoretical analysis shows that the mistake bound of the proposed method, thanks to the aggressive update trials, is better than conservative competitor in expectation. We finally empirically evaluate it on several real-world graph databases. Encouraging experimental results further demonstrate the effectiveness of our method.

Traditional online learning algorithms are designed for vector data only, which assume that the labels of all the training examples are provided. In this paper, we study graph classification where only limited nodes are chosen for labelling by selective sampling. Particularly, we first adapt a spectral-based graph regularization technique to derive a novel online learning linear algorithm which can handle graph data, although it still queries the labels of all nodes and thus is not preferred, as labelling is typically time-consuming. To address this issue, we then propose a new confidence-based query method for selective sampling. The theoretical result shows that our online learning algorithm with a fraction of queried labels can achieve a mistake bound comparable with the one learning on all labels of the nodes. In addition, the algorithm based on our proposed query strategy can achieve a mistake bound better than the one based on other query methods. However, our algorithm is conservative to update the model whenever error happens, which obviously wastes training labels that are valuable for the model. To take advantage of these labels, we further propose a novel aggressive algorithm, which can update the model aggressively even if no error occurs. The theoretical analysis shows that our aggressive approach can achieve a mistake bound better than its conservative and fully-supervised counterpart, with substantially fewer queried times. We empirically evaluate our algorithm on several real-world graph datasets and the experimental results demonstrate that our method is highly effective.

For many data mining and machine learning tasks, the quality of a similarity measure is the key for their performance. To automatically find a good similarity measure from datasets, metric learning and similarity learning are proposed and studied extensively. Metric learning will learn a Mahalanobis distance based on positive semi-definite (PSD) matrix, to measure the distances between objectives, while similarity learning aims to directly learn a similarity function without PSD constraint so that it is more attractive. Most of the existing similarity learning algorithms are online similarity learning method, since online learning is more scalable than offline learning. However, most existing online similarity learning algorithms learn a full matrix with d2parameters, where d is the dimension of the instances. This is clearly inefficient for high dimensional tasks due to its high memory and computational complexity. To solve this issue, we introduce several Sparse Online Relative Similarity (SORS) learning algorithms, which learn a sparse model during the learning process, so that the memory and computational cost can be significantly reduced. We theoretically analyze the proposed algorithms, and evaluate them on some real-world high dimensional datasets. Encouraging empirical results demonstrate the advantages of our approach in terms of efficiency and efficacy.

Cost-Sensitive Online Classification is recently proposed to directly online optimize two well-known cost-sensitive measures: (i) maximization of weighted sum of sensitivity and specificity, and (ii) minimization of weighted misclassification cost. However, the previous existing learning algorithms only utilized the first order information of the data stream. This is insufficient, as recent studies have proved that incorporating second order information could yield significant improvements on the prediction model. Hence, we propose a novel cost-sensitive online classification algorithm with adaptive regularization. We theoretically analyzed the proposed algorithm and empirically validated its effectiveness with extensive experiments. We also demonstrate the application of the proposed technique for solving several online anomaly detection tasks, showing that the proposed technique could be an effective tool to tackle cost-sensitive online classification tasks in various application domains.

The Alternating Direction Method of Multipliers (ADMM) has been studied for years. Traditional ADMM algorithms need to compute, at each iteration, an (empirical) expected loss function on all training examples, resulting in a computational complexity proportional to the number of training examples. To reduce the complexity, stochastic ADMM algorithms were proposed to replace the expected loss function with a random loss function associated with one uniformly drawn example plus a Bregman divergence term. The Bregman divergence, however, is derived from a simple 2nd-order proximal function, i.e., the half squared norm, which could be a suboptimal choice. In this paper, we present a new family of stochastic ADMM algorithms with optimal 2nd-order proximal functions, which produce a new family of adaptive stochastic ADMM methods. We theoretically prove that the regret bounds are as good as the bounds which could be achieved by the best proximal function that can be chosen in hindsight. Encouraging empirical results on a variety of real-world datasets confirm the effectiveness and efficiency of the proposed algorithms.

Uniform sampling of training data has been commonly used in traditional stochastic optimization algorithms such as Proximal Stochastic Mirror Descent (prox-SMD) and Proximal Stochastic Dual Coordinate Ascent (prox-SDCA). Although uniform sampling can guarantee that the sampled stochastic quantity is an unbiased estimate of the corresponding true quantity, the resulting estimator may have a rather high variance, which negatively affects the convergence of the underlying optimization procedure. In this paper we study stochastic optimization, including prox-SMD and prox-SDCA, with importance sampling, which improves the convergence rate by reducing the stochastic variance. We theoretically analyze the algorithms and empirically validate their effectiveness.

Learning relative similarity from pairwise instances is an important problem in machine learning and has a wide range of applications. Despite being studied for years, some existing methods solved by Stochastic Gradient Descent (SGD) techniques generally suffer from slow convergence. In this paper, we investigate the application of Stochastic Dual Coordinate Ascent (SDCA) technique to tackle the optimization task of relative similarity learning by extending from vector to matrix parameters. Theoretically, we prove the optimal linear convergence rate for the proposed SDCA algorithm, beating the well-known sublinear convergence rate by the previous best metric learning algorithms. Empirically, we conduct extensive experiments on both standard and large-scale data sets to validate the effectiveness of the proposed algorithm for retrieval tasks.

The amount of data in our society has been exploding in the era of big data today. In this paper, we address several open challenges of big data stream classification, including high volume, high velocity, high dimensionality, and high sparsity. Many existing studies in data mining literature solve data stream classification tasks in a batch learning setting, which suffers from poor efficiency and scalability when dealing with big data. To overcome the limitations, this paper investigates an online learning framework for big data stream classification tasks. Unlike some existing online data stream classification techniques that are often based on first-order online learning, we propose a framework of Sparse Online Classification (SOC) for data stream classification, which includes some state-of-the-art first-order sparse online learning algorithms as special cases and allows us to derive a new effective second-order online learning algorithm for data stream classification. We conduct an extensive set of experiments, in which encouraging results validate the efficacy of the proposed algorithms in comparison to a family of state-of-the-art techniques on a variety of data stream classification tasks.

Recent years have witnessed a number of studies on distance metric learning to improve visual similarity search in content-based image retrieval (CBIR). Despite their successes, most existing methods on distance metric learning are limited in two aspects. First, they usually assume the target proximity function follows the family of Mahalanobis distances, which limits their capacity of measuring similarity of complex patterns in real applications. Second, they often cannot effectively handle the similarity measure of multimodal data that may originate from multiple resources. To overcome these limitations, this paper investigates an online kernel similarity learning framework for learning kernel-based proximity functions which goes beyond the conventional linear distance metric learning approaches. Based on the framework, we propose a novel online multiple kernel similarity (OMKS) learning method which learns a flexible nonlinear proximity function with multiple kernels to improve visual similarity search in CBIR. We evaluate the proposed technique for CBIR on a variety of image data sets in which encouraging results show that OMKS outperforms the state-of-the-art techniques significantly.

Both cost-sensitive classification and online learning have been extensively studied in data mining and machine learning communities, respectively. However, very limited study addresses an important intersecting problem, that is, “Cost-Sensitive Online Classification". In this paper, we formally study this problem, and propose a new framework for Cost-Sensitive Online Classification by directly optimizing cost-sensitive measures using online gradient descent techniques. Specifically, we propose two novel cost-sensitive online classification algorithms, which are designed to directly optimize two well-known cost-sensitive measures: (i) maximization of weighted sum of sensitivity and specificity, and (ii) minimization of weighted misclassification cost. We analyze the theoretical bounds of the cost-sensitive measures made by the proposed algorithms, and extensively examine their empirical performance on a variety of cost-sensitive online classification tasks. Finally, we demonstrate the application of the proposed technique for solving several online anomaly detection tasks, showing that the proposed technique could be a highly efficient and effective tool to tackle cost-sensitive online classification tasks in various application domains.

Feature selection is an important technique for data mining. Despite its importance, most studies of feature selection are restricted to batch learning. Unlike traditional batch learning methods, online learning represents a promising family of efficient and scalable machine learning algorithms for large-scale applications. Most existing studies of online learning require accessing all the attributes/features of training instances. Such a classical setting is not always appropriate for real-world applications when data instances are of high dimensionality or it is expensive to acquire the full set of attributes/features. To address this limitation, we investigate the problem of online feature selection (OFS) in which an online learner is only allowed to maintain a classifier involved only a small and fixed number of features. The key challenge of online feature selection is how to make accurate prediction for an instance using a small number of active features. This is in contrast to the classical setup of online learning where all the features can be used for prediction. We attempt to tackle this challenge by studying sparsity regularization and truncation techniques. Specifically, this article addresses two different tasks of online feature selection: 1) learning with full input, where an learner is allowed to access all the features to decide the subset of active features, and 2) learning with partial input, where only a limited number of features is allowed to be accessed for each instance by the learner. We present novel algorithms to solve each of the two problems and give their performance analysis. We evaluate the performance of the proposed algorithms for online feature selection on several public data sets, and demonstrate their applications to real-world problems including image classification in computer vision and microarray gene expression analysis in bioinformatics. The encouraging results of our experiments validate the efficacy and efficiency of the proposed techniques.

Malicious Uniform Resource Locator (URL) detection is an important problem in web search and mining, which plays a critical role in internet security. In literature, many existing studies have attempted to formulate the problem as a regular supervised binary classification task, which typically aims to optimize the prediction accuracy. However, in a real-world malicious URL detection task, the ratio between the number of malicious URLs and legitimate URLs is highly imbalanced, making it very inappropriate for simply optimizing the prediction accuracy. Besides, another key limitation of the existing work is to assume a large amount of training data is available, which is impractical as the human labeling cost could be potentially quite expensive. To solve these issues, in this paper, we present a novel framework of Cost-Sensitive Online Active Learning (CSOAL), which only queries a small fraction of training data for labeling and directly optimizes two cost-sensitive measures to address the class-imbalance issue. In particular, we propose two CSOAL algorithms and analyze their theoretical performance in terms of cost-sensitive bounds. We conduct an extensive set of experiments to examine the empirical performance of the proposed algorithms for a large-scale challenging malicious URL detection task, in which the encouraging results showed that the proposed technique by querying an extremely small-sized labeled data (about 0.5% out of 1-million instances) can achieve better or highly comparable classification performance in comparison to the state-of-the-art cost-insensitive and cost-sensitive online classification algorithms using a huge amount of labeled data.

Recent years have witnessed extensive studies on distance metric learning (DML) for improving similarity search in multimedia information retrieval tasks. Despite their successes, most existing DML methods suffer from two critical limitations: (i) they typically attempt to learn a linear distance function on the input feature space, in which the assumption of linearity limits their capacity of measuring the similarity on complex patterns in real-world applications; (ii) they are often designed for learning distance metrics on uni-modal data, which may not effectively handle the similarity measures for multimedia objects with multimodal representations. To address these limitations, in this paper, we propose a novel framework of online multimodal deep similarity learning (OMDSL), which aims to optimally integrate multiple deep neural networks pretrained with stacked denoising autoencoder. In particular, the proposed framework explores a unified two-stage online learning scheme that consists of (i) learning a flexible nonlinear transformation function for each individual modality, and (ii) learning to find the optimal combination of multiple diverse modalities simultaneously in a coherent process. We conduct an extensive set of experiments to evaluate the performance of the proposed algorithms for multimodal image retrieval tasks, in which the encouraging results validate the effectiveness of the proposed technique.

Traditional batch model-based Collaborative Filtering (CF) approaches typically assume a collection of users' rating data is given a priori for training the model. They suffer from a common yet critical drawback, i.e., the model has to be re-trained completely from scratch whenever new training data arrives, which is clearly non-scalable for large real recommender systems where users' rating data often arrives sequentially and frequently. In this paper, we investigate a novel efficient and scalable online collaborative filtering technique for on-the-fly recommender systems, which is able to effectively online update the recommendation model from a sequence of rating observations. Specifically, we propose a family of online multi-task collaborative filtering (OMTCF) algorithms, which tackle the online collaborative filtering task by exploiting the similar principle as online multitask learning. Encouraging empirical results on large-scale datasets showed that the proposed technique is significantly more effective than the state-of-the-art algorithms.

Although both cost-sensitive classification and online learning have been well studied separately in data mining and machine learning, there was very few comprehensive study of cost-sensitive online classification in literature. In this paper, we formally investigate this problem by directly optimizing cost-sensitive measures for an online classification task. As the first comprehensive study, we propose the Cost-Sensitive Double Updating Online Learning (CSDUOL) algorithms, which explores a recent double updating technique to tackle the online optimization task of cost-sensitive classification by maximizing the weighted sum or minimizing the weighted misclassification cost. We theoretically analyze the cost-sensitive measure bounds of the proposed algorithms, extensively examine their empirical performance for cost-sensitive online classification tasks, and finally demonstrate the application of our technique to solve online anomaly detection tasks.

Online portfolio selection has been attracting increasing attention from the data mining and machine learning communities. All existing online portfolio selection strategies focus on the first order information of a portfolio vector, though the second order information may also be beneficial to a strategy. Moreover, empirical evidence shows that relative stock prices may follow the mean reversion property, which has not been fully exploited by existing strategies. This article proposes a novel online portfolio selection strategy named Confidence Weighted Mean Reversion (CWMR). Inspired by the mean reversion principle in finance and confidence weighted online learning technique in machine learning, CWMR models the portfolio vector as a Gaussian distribution, and sequentially updates the distribution by following the mean reversion trading principle. CWMR’s closed-form updates clearly reflect the mean reversion trading idea. We also present several variants of CWMR algorithms, including a CWMR mixture algorithm that is theoretical universal. Empirically, CWMR strategy is able to effectively exploit the power of mean reversion for online portfolio selection. Extensive experiments on various real markets show that the proposed strategy is superior to the state-of-the-art techniques. The experimental testbed including source codes and data sets is available online.

Both cost-sensitive classification and online learning have been studied extensively in data mining and machine learning communities, respectively. It is a bit surprising that there was very limited comprehensive study for addressing an important intersecting problem, that is, cost-sensitive online classification. In this paper, we formally study this problem, and propose a new framework for cost-sensitive online classification by exploiting the idea of online gradient descent techniques. Based on the framework, we propose a family of cost-sensitive online classification algorithms, which are designed to directly optimize two well-known cost-sensitive measures: (i) maximization of weighted sum of sensitivity and specificity, and (ii) minimization of weighted misclassification cost. We analyze the theoretical bounds of the cost-sensitive measures made by the proposed algorithms, and extensively examine their empirical performance on a variety of cost-sensitive online classification tasks.

This article proposes a novel online portfolio selection strategy named “Passive Aggressive Mean Reversion” (PAMR). Unlike traditional trend following approaches, the proposed approach relies upon the mean reversion relation of financial markets. Equipped with online passive aggressive learning technique from machine learning, the proposed portfolio selection strategy can effectively exploit the mean reversion property of markets. By analyzing PAMR’s update scheme, we find that it nicely trades off between portfolio return and volatility risk and reflects the mean reversion trading principle. We also present several variants of PAMR algorithm, including a mixture algorithm which mixes PAMR and other strategies. We conduct extensive numerical experiments to evaluate the empirical performance of the proposed algorithms on various real datasets. The encouraging results show that in most cases the proposed PAMR strategy outperforms all benchmarks and almost all state-of-the-art portfolio selection strategies under various performance metrics. In addition to its superior performance, the proposed PAMR runs extremely fast and thus is very suitable for real-life online trading applications. The experimental testbed including source codes and data sets is available at http://www.cais.ntu.edu.sg/~chhoi/PAMR/ .

With the popularity of various social media applications, massive social images associated with high quality tags have been made available in many social media web sites nowadays. Mining social images on the web has become an emerging important research topic in web search and data mining. In this paper, we propose a machine learning framework for mining social images and investigate its application to automated image tagging. To effectively discover knowledge from social images that are often associated with multimodal contents (including visual images and textual tags), we propose a novel Unified Distance Metric Learning (UDML) scheme, which not only exploits both visual and textual contents of social images, but also effectively unifies both inductive and transductive metric learning techniques in a systematic learning framework. We further develop an efficient stochastic gradient descent algorithm for solving the UDML optimization task and prove the convergence of the algorithm. By applying the proposed technique to the automated image tagging task in our experiments, we demonstrate that our technique is empirically effective and promising for mining social images towards some real applications.

Sparse coding which encodes the original signal in a sparse signal space, has shown its state-of-the-art performance in the visual codebook generation and feature quantization process of BoW based image representation. However, in the feature quantization process of sparse coding, some similar local features may be quantized into different visual words of the codebook due to the sensitiveness of quantization. In this paper, to alleviate the impact of this problem, we propose a Laplacian sparse coding method, which will exploit the dependence among the local features. Specifically, we propose to use histogram intersection based kNN method to construct a Laplacian matrix, which can well characterize the similarity of local features. In addition, we incorporate this Laplacian matrix into the objective function of sparse coding to preserve the consistence in sparse representation of similar local features. Comprehensive experimental results show that our method achieves or outperforms existing state-of-the-art results, and exhibits excellent performance on Scene 15 data set.

In most online learning algorithms, the weights assigned to the misclassified examples (or support vectors) remain unchanged during the entire learning process. This is clearly insufficient since when a new misclassified example is added to the pool of support vectors, we generally expect it to affect the weights for the existing support vectors. In this paper, we propose a new online learning method, termed Double Updating Online Learning", or "DUOL" for short. Instead of only assigning a fixed weight to the misclassified example received in current trial, the proposed online learning algorithm also tries to update the weight for one of the existing support vectors. We show that the mistake bound can be significantly improved by the proposed online learning method. Encouraging experimental results show that the proposed technique is in general considerably more effective than the state-of-the-art online learning algorithms."