Recent studies have raised significant concerns regarding the reliability of current mathematical benchmarks, highlighting key limitations such as simplistic design and potential data contamination that undermine evaluation accuracy. Consequently, developing a reliable benchmark that effectively evaluates large language models' (LLMs) genuine capabilities in mathematical reasoning remains a critical challenge. To address these concerns, we propose RV-Bench, a novel evaluation methodology for Benchmarking LLMs with Random Variables in mathematical reasoning. Specifically, we develop question-generating functions to produce random variable questions (RVQs), whose background content mirrors the original benchmark problems, but with randomized variable combinations, rendering them "unseen" to LLMs. Models must completely understand the inherent question pattern to correctly answer RVQs with diverse variable combinations. Thus, an LLMs' genuine reasoning capability is reflected through its accuracy and robustness on RV-Bench. We conducted extensive experiments on over 30 representative LLMs across more than 1,000 RVQs. Our findings reveal that LLMs exhibit a proficiency imbalance between encountered and "unseen" data distributions. Furthermore, RV-Bench reveals that proficiency generalization across similar mathematical reasoning tasks is limited, but we verified that it can still be effectively elicited through test-time scaling.

The emergence of Multimodal Large Language Models (MLLMs) has propelled the development of autonomous agents that operate on Graphical User Interfaces (GUIs) using pure visual input. A fundamental challenge is robustly grounding natural language instructions. This requires a precise spatial alignment, which accurately locates the coordinates of each element, and, more critically, a correct semantic alignment, which matches the instructions to the functionally appropriate UI element. Although Reinforcement Learning with Verifiable Rewards (RLVR) has proven to be effective at improving spatial alignment for these MLLMs, we find that inefficient exploration bottlenecks semantic alignment, which prevents models from learning difficult semantic associations. To address this exploration problem, we present Adaptive Exploration Policy Optimization (AEPO), a new policy optimization framework. AEPO employs a multi-answer generation strategy to enforce broader exploration, which is then guided by a theoretically grounded Adaptive Exploration Reward (AER) function derived from first principles of efficiency η=U/C. Our AEPO-trained models, InfiGUI-G1-3B and InfiGUI-G1-7B, establish new state-of-the-art results across multiple challenging GUI grounding benchmarks, achieving significant relative improvements of up to 9.0% against the naive RLVR baseline on benchmarks designed to test generalization and semantic understanding.

To tackle increasingly complex tasks, recent research on mobile agents has shifted towards multi-agent collaboration. Current mobile multi-agent systems are primarily deployed in the cloud, leading to high latency and operational costs. A straightforward idea is to deploy a device–cloud collaborative multi-agent system, which is nontrivial, as directly extending existing systems introduces new challenges: (1) reliance on cloud-side verification requires uploading mobile screenshots, compromising user privacy; and (2) open-loop cooperation lacking device-to-cloud feedback, underutilizing device resources and increasing latency. To overcome these limitations, we propose EcoAgent, a closed-loop device-cloud collaborative multi-agent framework designed for privacy-aware, efficient, and responsive mobile automation. EcoAgent integrates a novel reasoning approach, Dual-ReACT, into the cloud-based Planning Agent, fully exploiting cloud reasoning to compensate for limited on-device capacity, thereby enabling device-side verification and lightweight feedback. Furthermore, the device-based Observation Agent leverages a Pre-understanding Module to summarize screen content into concise textual descriptions, significantly reducing token usage and device-cloud communication overhead while preserving privacy. Experiments on AndroidWorld demonstrate that EcoAgent matches the task success rates of fully cloud-based agents, while reducing resource consumption and response latency.

Xueyu Hu, Tao Xiong, Biao Yi, Zishu Wei, Ruixuan Xiao, Yurun Chen, Jiasheng Ye, Meiling Tao, Xiangxin Zhou, Ziyu Zhao, Yuhuai Li, Shengze Xu, Shenzhi Wang, Xinchen Xu, Shuofei Qiao, Zhaokai Wang, Kun Kuang, Tieyong Zeng, Liang Wang, Jiwei Li, Yuchen Eleanor Jiang, Wangchunshu Zhou, Guoyin Wang, Keting Yin, Zhou Zhao, Hongxia Yang, Fan Wu, Shengyu Zhang, Fei Wu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Haotian Zhou, Tingkai Liu, Qianli Ma, Yufeng Zhang, Jianbo Yuan, Pengfei Liu, Yang You, Hongxia Yang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Extrapolating ultra-long contexts (text length $>$128K) remains a major challenge for large language models (LLMs), as most training-free extrapolation methods are not only severely limited by memory bottlenecks, but also suffer from the attention sink, which restricts their scalability and effectiveness in practice. In this work, we propose ParallelComp, a parallel long-context compression method that effectively overcomes the memory bottleneck, enabling 8B-parameter LLMs to extrapolate from 8K to 128K tokens on a single A100 80GB GPU in a training-free setting. ParallelComp splits the input into chunks, dynamically evicting redundant chunks and irrelevant tokens, supported by a parallel KV cache eviction mechanism. Importantly, we present a systematic theoretical and empirical analysis of attention biases in parallel attention—including the attention sink, recency bias, and middle bias—and reveal that these biases exhibit distinctive patterns under ultra-long context settings. We further design a KV cache eviction technique to mitigate this phenomenon. Experimental results show that ParallelComp enables an 8B model (trained on 8K context) to achieve 91.17% of GPT-4’s performance under ultra-long contexts, outperforming closed-source models such as Claude-2 and Kimi-Chat. We achieve a 1.76x improvement in chunk throughput, thereby achieving a 23.50x acceleration in the prefill stage with negligible performance loss and pave the way for scalable and robust ultra-long contexts extrapolation in LLMs. We release the code at https://github.com/menik1126/ParallelComp.

Foundation models (FMs), known for their broad cognitive capabilities but often constrained to cloud deployment, and specialized models (SMs), characterized by their lightweight, goal-oriented nature suitable for devices, offer complementary strengths. Traditional cloud-centric paradigms face limitations in real-time performance, personalization, cost, and privacy, highlighting the need for innovative approaches that leverage device-level capabilities. This workshop served as a platform to discuss the rapid advancements and emerging research directions in FM-SM collaboration and co-evolution. Key focus areas included: (i) novel collaborative frameworks bridging cloud FMs and device SMs, (ii) mechanisms for model evolution, knowledge transfer, aggregation, and generation, (iii) integration of multimodal perspectives, particularly for multimedia retrieval tasks relevant to ICMR, (iv) strategies for enhancing robustness, interpretability, and fairness, and (v) the development of new benchmarks and resources. Featuring keynote presentations and peer-reviewed papers on topics ranging from multimodal understanding and reasoning to efficient on-device fine-tuning and mobile agents, the workshop fostered interdisciplinary dialogue.

Model fusion combines multiple Large Language Models (LLMs) with different strengths into a more powerful, integrated model through lightweight training methods. Existing works on model fusion focus primarily on supervised fine-tuning (SFT), leaving preference alignment (PA) —a critical phase for enhancing LLM performance—largely unexplored. The current few fusion methods on PA phase, like WRPO, simplify the process by utilizing only response outputs from source models while discarding their probability information. To address this limitation, we propose InfiFPO, a preference optimization method for implicit model fusion. InfiFPO replaces the reference model in Direct Preference Optimization (DPO) with a fused source model that synthesizes multi-source probabilities at the sequence level, circumventing complex vocabulary alignment challenges in previous works and meanwhile maintaining the probability information. By introducing probability clipping and max-margin fusion strategies, InfiFPO enables the pivot model to align with human preferences while effectively distilling knowledge from source models. Comprehensive experiments on 11 widely-used benchmarks demonstrate that InfiFPO consistently outperforms existing model fusion and preference optimization methods. When using Phi-4 as the pivot model, InfiFPO improves its average performance from 79.95 to 83.33 on 11 benchmarks, significantly improving its capabilities in mathematics, coding, and reasoning tasks.

Recent advances in large language models (LLMs) have intensified efforts to fuse heterogeneous open-source models into a unified system that inherits their complementary strengths. Existing logit-based fusion methods maintain inference efficiency but treat vocabulary dimensions independently, overlooking semantic dependencies encoded by cross-dimension interactions. These dependencies reflect how token types interact under a model's internal reasoning and are essential for aligning models with diverse generation behaviors. To explicitly model these dependencies, we propose \textbf{InfiGFusion}, the first structure-aware fusion framework with a novel \textit{Graph-on-Logits Distillation} (GLD) loss. Specifically, we retain the top-$k$ logits per output and aggregate their outer products across sequence positions to form a global co-activation graph, where nodes represent vocabulary channels and edges quantify their joint activations. To ensure scalability and efficiency, we design a sorting-based closed-form approximation that reduces the original $O(n^4)$ cost of Gromov-Wasserstein distance to $O(n \log n)$, with provable approximation guarantees. Experiments across multiple fusion settings show that GLD consistently improves fusion quality and stability. InfiGFusion outperforms SOTA models and fusion baselines across 11 benchmarks spanning reasoning, coding, and mathematics. It shows particular strength in complex reasoning tasks, with +35.6 improvement on Multistep Arithmetic and +37.06 on Causal Judgement over SFT, demonstrating superior multi-step and relational inference.

Graph Neural Networks (GNNs) show promising results for graph tasks. However, existing GNNs' generalization ability will degrade when there exist distribution shifts between testing and training graph data. The fundamental reason for the severe degeneration is that most GNNs are designed based on the I.I.D hypothesis. In such a setting, GNNs tend to exploit subtle statistical correlations existing in the training set for predictions, even though it is a spurious correlation. In this paper, we study the problem of the generalization ability of GNNs on Out-Of-Distribution (OOD) settings. To solve this problem, we propose the Learning to Reweight for Generalizable Graph Neural Network (L2R-GNN) to enhance the generalization ability for achieving satisfactory performance on unseen testing graphs that have different distributions with training graphs. We propose a novel nonlinear graph decorrelation method, which can substantially improve the out-of-distribution generalization ability and compares favorably to previous methods in restraining the over-reduced sample size. The variables of graph representation are clustered based on the stability of their correlations, and graph decorrelation method learns weights to remove correlations between the variables of different clusters rather than any two variables. Besides, we introduce an effective stochastic algorithm based on bi-level optimization for the L2R-GNN framework, which enables simultaneously learning the optimal weights and GNN parameters, and avoids the over-fitting issue. Experiments show that L2R-GNN greatly outperforms baselines on various graph prediction benchmarks under distribution shifts.

Ziwei Chai, Guoyin Wang, Jing Su, Tianjie Zhang, Xuanwen Huang, Xuwu Wang, Jingjing Xu, Jianbo Yuan, Hongxia Yang, Fei Wu, Yang Yang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Yiren Jian, Tingkai Liu, Yunzhe Tao, Chunhui Zhang, Soroush Vosoughi, Hongxia Yang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Tingkai Liu, Yunzhe Tao, Haogeng Liu, Qihang Fang, Ding Zhou, Huaibo Huang, Ran He, Hongxia Yang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Discussion and debate among Large Language Models (LLMs) have gained considerable attention due to their potential to enhance the reasoning ability of LLMs. Although natural language is an obvious choice for communication due to LLM's language understanding capability, the token sampling step needed when generating natural language poses a potential risk of information loss, as it uses only one token to represent the model's belief across the entire vocabulary. In this paper, we introduce a communication regime named CIPHER (Communicative Inter-Model Protocol Through Embedding Representation) to address this issue. Specifically, we remove the token sampling step from LLMs and let them communicate their beliefs across the vocabulary through the expectation of the raw transformer output embeddings. Remarkably, by deviating from natural language, CIPHER offers an advantage of encoding a broader spectrum of information without any modification to the model weights, outperforming the state-of-the-art LLM debate methods using natural language by 0.5-5.0% across five reasoning tasks and multiple open-source LLMs of varying sizes. This showcases the superiority and robustness of embeddings as an alternative "language" for communication among LLMs. We anticipate that CIPHER will inspire further exploration for the design of interactions within LLM agent systems, offering a new direction that could significantly influence future developments in the field.

Scaling of deep neural networks, especially Transformers, is pivotal for their surging performance and has further led to the emergence of sophisticated reasoning capabilities in foundation models. Such scaling generally requires training large models from scratch with random initialization, failing to leverage the knowledge acquired by their smaller counterparts, which are already resource-intensive to obtain. To tackle this inefficiency, we present $\textbf{L}$ossl$\textbf{E}$ss $\textbf{MO}$del Expansio$\textbf{N}$ (LEMON), a recipe to initialize scaled models using the weights of their smaller but pre-trained counterparts. This is followed by model training with an optimized learning rate scheduler tailored explicitly for the scaled models, substantially reducing the training time compared to training from scratch. Notably, LEMON is versatile, ensuring compatibility with various network structures, including models like Vision Transformers and BERT. Our empirical results demonstrate that LEMON reduces computational costs by 56.7\% for Vision Transformers and 33.2\% for BERT when compared to training from scratch.

Diffusion models excel at generating photo-realistic images but come with significant computational costs in both training and sampling. While various techniques address these computational challenges, a less-explored issue is designing an efficient and adaptable network backbone for iterative refinement. Current options like U-Net and Vision Transformer often rely on resource-intensive deep networks and lack the flexibility needed for generating images at variable resolutions or with a smaller network than used in training. This study introduces LEGO bricks, which seamlessly integrate Local-feature Enrichment and Global-content Orchestration. These bricks can be stacked to create a test-time reconfigurable diffusion backbone, allowing selective skipping of bricks to reduce sampling costs and generate higher-resolution images than the training data. LEGO bricks enrich local regions with an MLP and transform them using a Transformer block while maintaining a consistent full-resolution image across all bricks. Experimental results demonstrate that LEGO bricks enhance training efficiency, expedite convergence, and facilitate variable-resolution image generation while maintaining strong generative performance. Moreover, LEGO significantly reduces sampling time compared to other methods, establishing it as a valuable enhancement for diffusion models. Our code and project page are available at https://jegzheng.github.io/LEGODiffusion.

Multi-modal Large Language Models (MLLMs) are increasingly prominent in the field of artificial intelligence. Visual instruction fine-tuning (IFT) is a vital process for aligning MLLMs’ output with user’s intentions. High-quality and diversified instruction following data is the key to this fine-tuning process. Recent studies propose to construct visual IFT datasets through a multifaceted approach: transforming existing datasets with rule-based templates, employing GPT-4 for rewriting annotations, and utilizing GPT-4V for visual dataset pseudo-labeling. LLaVA-1.5 adopted similar approach and construct LLaVA-Instruct, which is one of the simplest, most widely used, yet most effective IFT datasets today. Notably, when properly fine-tuned with this dataset, MLLMs can achieve state-of-the-art performance on several benchmarks. However, we noticed that models trained with this dataset often struggle to follow user instructions properly in multi-round dialog. In addition, tradition caption and VQA evaluation benchmarks, with their closed-form evaluation structure, are not fully equipped to assess the capabilities of modern open-ended generative MLLMs. This problem is not unique to the LLaVA-Instruct dataset, but may be a potential issue in all IFT datasets constructed from image captioning or VQA sources, though the extent of this issue may vary. We argue that datasets with diverse and high-quality detailed instruction following annotations are essential and adequate for MLLMs IFT. In this work, we establish a new IFT dataset, with images sourced from the COCO dataset along with more diverse instructions. Our experiments show that when fine-tuned with proposed dataset, MLLMs achieve better performance on open-ended evaluation benchmarks in both single-round and multi-round dialog setting.

In this work, we leverage the intrinsic segmentation of language sequences and design a new positional encoding method called Bilevel Positional Encoding (BiPE). For each position, our BiPE blends an intra-segment encoding and an inter-segment encoding. The intra-segment encoding identifies the locations within a segment and helps the model capture the semantic information therein via absolute positional encoding. The inter-segment encoding specifies the segment index, models the relationships between segments, and aims to improve extrapolation capabilities via relative positional encoding. Theoretical analysis shows this disentanglement of positional information makes learning more effective. The empirical results also show that our BiPE has superior length extrapolation capabilities across a wide range of tasks in diverse text modalities.

In this paper, we introduce InfiAgent-DABench, the first benchmark specifically designed to evaluate LLM-based agents on data analysis tasks. Agents need to solve these tasks end-to-end by interacting with an execution environment. This benchmark contains DAEval, a dataset consisting of 603 data analysis questions derived from 124 CSV files, and an agent framework which incorporates LLMs to serve as data analysis agents for both serving and evaluating. Since data analysis questions are often open-ended and hard to evaluate without human supervision, we adopt a format-prompting technique to convert each question into a closed-form format so that they can be automatically evaluated. Our extensive benchmarking of 34 LLMs uncovers the current challenges encountered in data analysis tasks. In addition, building upon our agent framework, we develop a specialized agent, DAAgent, which surpasses GPT-3.5 by 3.9% on DABench. Evaluation datasets and toolkits for InfiAgent-DABench are released at https://github.com/InfiAgent/InfiAgent.

In this work, we introduce self-infilling code generation, a general framework that incorporates infilling operations into auto-regressive decoding. Our approach capitalizes on the observation that recent infilling-capable code language models can perform self-infilling: whereas conventional infilling is designed to fill in the middle based on a predefined prefix and suffix, self-infilling sequentially generates both such surrounding context and the infilled content. We utilize self-infilling to introduce novel interruption and looping mechanisms in conventional decoding, evolving it into a non-monotonic process. Interruptions allow for postponing the generation of specific code until a definitive suffix is established, enhancing control during decoding. Meanwhile, the looping mechanism, which leverages the complementary nature of self-infilling and left-to-right decoding, can iteratively update and synchronize each piece of generation cyclically. Extensive experiments across a variety of code generation benchmarks demonstrate that decoding with self-infilling not only improves the output quality but also regularizes the overall generation, which effectively mitigates potential degeneration and scaffolds code to be more consistent with intended functionality.

Large Language Models for code (code LLMs) have witnessed tremendous progress in recent years. With the rapid development of code LLMs, many popular evaluation benchmarks, such as HumanEval, DS-1000, and MBPP, have emerged to measure the performance of code LLMs with a particular focus on code generation tasks. However, they are insufficient to cover the full range of expected capabilities of code LLMs, which span beyond code generation to answering diverse coding-related questions. To fill this gap, we propose InfiBench, the first large-scale freeform question-answering (QA) benchmark for code to our knowledge, comprising 234 carefully selected high-quality Stack Overflow questions that span across 15 programming languages. InfiBench uses four types of model-free automatic metrics to evaluate response correctness where domain experts carefully concretize the criterion for each question. We conduct a systematic evaluation for over 100 latest code LLMs on InfiBench, leading to a series of novel and insightful findings. Our detailed analyses showcase potential directions for further advancement of code LLMs. InfiBench is fully open source at https://infi-coder.github.io/infibench and continuously expanding to foster more scientific and systematic practices for code LLM evaluation.

Image restoration (IR) in real-world scenarios presents significant challenges due to the lack of high-capacity models and comprehensive datasets. To tackle these issues, we present a dual strategy: GenIR, an innovative data curation pipeline, and DreamClear, a cutting-edge Diffusion Transformer (DiT)-based image restoration model. **GenIR**, our pioneering contribution, is a dual-prompt learning pipeline that overcomes the limitations of existing datasets, which typically comprise only a few thousand images and thus offer limited generalizability for larger models. GenIR streamlines the process into three stages: image-text pair construction, dual-prompt based fine-tuning, and data generation \& filtering. This approach circumvents the laborious data crawling process, ensuring copyright compliance and providing a cost-effective, privacy-safe solution for IR dataset construction. The result is a large-scale dataset of one million high-quality images. Our second contribution, **DreamClear**, is a DiT-based image restoration model. It utilizes the generative priors of text-to-image (T2I) diffusion models and the robust perceptual capabilities of multi-modal large language models (MLLMs) to achieve photorealistic restoration. To boost the model's adaptability to diverse real-world degradations, we introduce the Mixture of Adaptive Modulator (MoAM). It employs token-wise degradation priors to dynamically integrate various restoration experts, thereby expanding the range of degradations the model can address. Our exhaustive experiments confirm DreamClear's superior performance, underlining the efficacy of our dual strategy for real-world image restoration. Code and pre-trained models are available at: https://github.com/shallowdream204/DreamClear.

In the realm of Multimodal Large Language Models (MLLMs), vision-language connector plays a crucial role to link the pre-trained vision encoders with Large Language Models (LLMs). Despite its importance, the vision-language connector has been relatively less explored. In this study, we aim to propose a strong vision-language connector that enables MLLM to simultaneously achieve high accuracy and low computation cost. We first reveal the existence of the visual anchors in Vision Transformer and propose a cost-effective search algorithm to progressively extract them. Building on these findings, we introduce the Anchor Former (AcFormer), a novel vision-language connector designed to leverage the rich prior knowledge obtained from these visual anchors during pretraining, guiding the aggregation of information. Through extensive experimentation, we demonstrate that the proposed method significantly reduces computational costs by nearly two-thirds, while simultaneously outperforming baseline methods. This highlights the effectiveness and efficiency of AcFormer.

Recommendation performance usually exhibits a long-tail distribution over users — a small portion of head users enjoy much more accurate recommendation services than the others. We reveal two sources of this performance heterogeneity problem: the uneven distribution of historical interactions (a natural source); and the biased training of recommender models (a model source). As addressing this problem cannot sacrifice the overall performance, a wise choice is to eliminate the model bias while maintaining the natural heterogeneity. The key to debiased training lies in eliminating the effect of confounders that influence both the user's historical behaviors and the next behavior. The emerging causal recommendation methods achieve this by modeling the causal effect between user behaviors, however potentially neglect unobserved confounders (e.g., friend suggestions) that are hard to measure in practice. To address unobserved confounders, we resort to the front-door adjustment (FDA) in causal theory and propose a causal multi-teacher distillation framework (CausalD). FDA requires proper mediators in order to estimate the causal effects of historical behaviors on the next behavior. To achieve this, we equip CausalD with multiple heterogeneous recommendation models to model the mediator distribution. Then, the causal effect estimated by FDA is the expectation of recommendation prediction over the mediator distribution and the prior distribution of historical behaviors, which is technically achieved by multi-teacher ensemble. To pursue efficient inference, CausalD further distills multiple teachers into one student model to directly infer the causal effect for making recommendations. We instantiate CausalD on two representative models, DeepFM and DIN, and conduct extensive experiments on three real-world datasets, which validate the superiority of CausalD over state-of-the-art methods. Through in-depth analysis, we find that CausalD largely improves the performance of tail users, reduces the performance heterogeneity, and enhances the overall performance.

In recommender systems, click behaviors play a fundamental role in mining users’ interests and training models (clicked items as positive samples). Such signals are implicit feedback and are arguably less representative of users’ inherent interests. Most existing works denoise implicit feedback by introducing external signals, such as gaze, dwell time, and “like” behaviors. However, such explicit feedback is not always routinely available, or might be problematic to collect on a large scale. In this paper, we identify that an interaction’s related structural patterns in its neighborhood graph are potentially correlated with some outcome of implicit feedback (i.e., users’ ratings after consuming items), analogous to findings in other domains such as social networks. Inspired by this finding, we propose a novel Structure LEarning based Denoising (SLED) framework for denoising recommendation without explicit signals, which consists of two phases: center-aware graph structure learning and denoised recommendation . Phase 1 pre-trains a structural encoder in a self-supervised manner and learns to capture an interaction’s related structural patterns in its neighborhood graph. Phase 2 transfers the structure encoder to downstream recommendation datasets, which helps to down-weight the effect of noisy interactions on user interest modeling and loss calculation. We collect a relatively noisy industrial dataset across several days during a period of product promotion festival. Extensive experiments on this dataset and multiple public datasets demonstrate that the proposed SLED framework can significantly improve the recommendation quality over various base recommendation models.

Contrastive learning-based vision-language pretraining approaches, such as CLIP, have demonstrated great success in many vision-language tasks. These methods achieve cross-modal alignment by encoding a matched image-text pair with similar feature embeddings, which are generated by aggregating information from visual patches and language tokens. However, direct aligning cross-modal information using such representations is challenging, as visual patches and text tokens differ in semantic levels and granularities. To alleviate this issue, we propose a Finite Discrete Tokens (FDT) based multimodal representation. FDT is a set of learnable tokens representing certain visualsemantic concepts. Both images and texts are embedded using shared FDT by first grounding multimodal inputs to FDT space and then aggregating the activated FDT representations. The matched visual and semantic concepts are enforced to be represented by the same set of discrete tokens by a sparse activation constraint. As a result, the granularity gap between the two modalities is reduced. Through both quantitative and qualitative analyses, we demonstrate that using FDT representations in CLIP-style models improves cross-modal alignment and performance in visual recognition and vision-language downstream tasks. Furthermore, we show that our method can learn more comprehensive representations, and the learned FDT capture meaningful cross-modal correspondence, ranging from objects to actions and attributes.11The source code can be found at https://github.com/yuxiaochen1103/FDT.

Graph neural networks (GNNs) have attracted tremendous attention from the graph learning community in recent years. It has been widely adopted in various real-world applications from diverse domains, such as social networks and biological graphs. The research and applications of graph deep learning present new challenges, including the sparse nature of graph data, complicated training of GNNs, and non-standard evaluation of graph tasks. To tackle the issues, we present CogDL1, a comprehensive library for graph deep learning that allows researchers and practitioners to conduct experiments, compare methods, and build applications with ease and efficiency. In CogDL, we propose a unified design for the training and evaluation of GNN models for various graph tasks, making it unique among existing graph learning libraries. By utilizing this unified trainer, CogDL can optimize the GNN training loop with several training techniques, such as mixed precision training. Moreover, we develop efficient sparse operators for CogDL, enabling it to become the most competitive graph library for efficiency. Another important CogDL feature is its focus on ease of use with the aim of facilitating open and reproducible research of graph learning. We leverage CogDL to report and maintain benchmark results on fundamental graph tasks, which can be reproduced and directly used by the community.

Device Model Generalization (DMG) is a practical yet under-investigated research topic for on-device machine learning applications. It aims to improve the generalization ability of pre-trained models when deployed on resource-constrained devices, such as improving the performance of pre-trained cloud models on smart mobiles. While quite a lot of works have investigated the data distribution shift across clouds and devices, most of them focus on model fine-tuning on personalized data for individual devices to facilitate DMG. Despite their promising, these approaches require on-device re-training, which is practically infeasible due to the overfitting problem and high time delay when performing gradient calculation on real-time data. In this paper, we argue that the computational cost brought by fine-tuning can be rather unnecessary. We consequently present a novel perspective to improving DMG without increasing computational cost, i.e., device-specific parameter generation which directly maps data distribution to parameters. Specifically, we propose an efficient Device-cloUd collaborative parametErs generaTion framework (DUET). DUET is deployed on a powerful cloud server that only requires the low cost of forwarding propagation and low time delay of data transmission between the device and the cloud. By doing so, DUET can rehearse the device-specific model weight realizations conditioned on the personalized real-time data for an individual device. Importantly, our DUET elegantly connects the cloud and device as a “duet” collaboration, frees the DMG from fine-tuning, and enables a faster and more accurate DMG paradigm. We conduct an extensive experimental study of DUET on three public datasets, and the experimental results confirm our framework’s effectiveness and generalisability for different DMG tasks.

In recommender systems, users' behavior data are driven by the interactions of user-item latent factors. To improve recommendation effectiveness and robustness, recent advances focus on latent factor disentanglement via variational inference. Despite significant progress, uncovering the underlying interactions, i.e., dependencies of latent factors, remains largely neglected by the literature. To bridge the gap, we investigate the joint disentanglement of user-item latent factors and the dependencies between them, namely latent structure learning. We propose to analyze the problem from the causal perspective, where a latent structure should ideally reproduce observational interaction data, and satisfy the structure acyclicity and dependency constraints, i.e., causal prerequisites. We further identify the recommendation-specific challenges for latent structure learning, i.e., the subjective nature of users' minds and the inaccessibility of private/sensitive user factors causing universally learned latent structure to be suboptimal for individuals. To address these challenges, we propose the personalized latent structure learning framework for recommendation, namely PlanRec, which incorporates 1) differentiable Reconstruction, Dependency, and Acyclicity regularizations to satisfy the causal prerequisites; 2) Personalized Structure Learning (PSL) which personalizes the universally learned dependencies through probabilistic modeling; and 3) uncertainty estimation which explicitly measures the uncertainty of structure personalization, and adaptively balances personalization and shared knowledge for different users. We conduct extensive experiments on two public benchmark datasets from MovieLens and Amazon, and a large-scale industrial dataset from Alipay. Empirical studies validate that PlanRec discovers effective shared/personalized structures, and successfully balances shared knowledge and personalization via rational uncertainty estimation.

Inferring new facts from an existing knowledge graph with explainable reasoning processes is an important problem, known as knowledge graph (KG) reasoning. The problem is often formulated as finding the specific path that represents the query relation and connects the query entity and the correct answer. However, due to the limited expressiveness of individual paths, the majority of previous works failed to capture the complex subgraph structure in the graph. We propose CogKR that traverses the knowledge graph to conduct multi-hop reasoning. More specifically, motivated by the dual process theory from cognitive science, our framework is composed of an extension module and a reasoning module. By setting up a cognitive graph through iteratively coordinating the two modules, CogKR can cope with more complex reasoning scenarios in the form of subgraphs instead of individual paths. Experiments on three knowledge graph reasoning benchmarks demonstrate that CogKR achieves significant improvements in accuracy compared with previous methods while providing the explainable capacity. Moreover, we evaluate CogKR on the challenging one-shot link prediction task, exhibiting the superiority of the framework on accuracy and scalability compared to the state-of-the-art approaches.

Graph-based recommendation systems are blossoming recently, which models user-item interactions as a user-item graph and utilizes graph neural networks (GNNs) to learn the embeddings for users and items. A fundamental challenge of graph-based recommendation is that there only exists observed positive user-item pairs in the user-item graph. Negative sampling is a vital technique to solve the one-class problem and is widely used in many recommendation methods. However, the previous works only focus on the design of negative sampling distribution but ignore the sampled region for negative sampling. In this work, we propose the Three-Region Principle to guide negative sampling, which suggests that we should negatively sample more items at an intermediate region and less adjacent and distant items. In light of this principle, we present the RecNS method, which is a general negative sampling method designed with two sampling strategies: positive-assisted sampling and exposure-augmented sampling. Instead of sampling existing negative items from graph data, we merge these two strategies in embedding space to generate negative item embeddings. Extensive experiments demonstrate that our proposed RecNS method significantly outperforms all negative sampling baselines, e.g., 10.47% for PinSage, 6.02% for NGCF, and 8.20% for LightGCN in terms of Recall@20 on the Alibaba dataset.

Influenced by the great success of deep learning via cloud computing and the rapid development of edge chips, research in artificial intelligence (AI) has shifted to both of the computing paradigms, i.e., cloud computing and edge computing. In recent years, we have witnessed significant progress in developing more advanced AI models on cloud servers that surpass traditional deep learning models owing to model innovations (e.g., Transformers, Pretrained families), explosion of training data and soaring computing capabilities. However, edge computing, especially edge and cloud collaborative computing, are still in its infancy to announce their success due to the resource-constrained IoT scenarios with very limited algorithms deployed. In this survey, we conduct a systematic review for both cloud and edge AI. Specifically, we are the first to set up the collaborative learning mechanism for cloud and edge modeling with a thorough review of the architectures that enable such mechanism. We also discuss potentials and practical experiences of some on-going advanced edge AI topics including pretraining models, graph neural networks and reinforcement learning. Finally, we discuss the promising directions and challenges in this field.

Feed recommendation allows users to constantly browse items until feel uninterested and leave the session, which differs from traditional recommendation scenarios. Within a session, user's decision to continue browsing or not substantially affects occurrences of later clicks. However, such type of exposure bias is generally ignored or not explicitly modeled in most feed recommendation studies. In this paper, we model this effect as part of intra-session context, and propose a novel intra-session Context-aware Feed Recommendation (INSCAFER) framework to maximize the total views and total clicks simultaneously. User click and browsing decisions are jointly learned by a multi-task setting, and the intra-session context is encoded by the session-wise exposed item sequence. We deploy our model on Alipay with all key business benchmarks improved. Our method sheds some lights on feed recommendation studies which aim to optimize session-level click and view metrics.

Despite the remarkable success of deep multi-modal learning in practice, it has not been well-explained in theory. Recently, it has been observed that the best uni-modal network outperforms the jointly trained multi-modal network across different combinations of modalities on various tasks, which is counter-intuitive since multiple signals would bring more information (Wang et al., 2020). This work provides a theoretical explanation for the emergence of such performance gap in neural networks for the prevalent joint training framework. Based on a simplified data distribution that captures the realistic property of multi-modal data, we prove that for multi-modal late-fusion network with (smoothed) ReLU activation trained jointly by gradient descent, different modalities will compete with each other and only a subset of modalities will be learned by its corresponding encoder networks. We refer to this phenomenon as modality competition, and the losing modalities, which fail to be discovered, are the origins where the sub-optimality of joint training comes from. In contrast, for uni-modal networks with similar learning settings, we provably show that the networks will focus on learning modality-associated features. Experimentally, we illustrate that modality competition matches the intrinsic behavior of late-fusion joint training to supplement our theoretical results. To the best of our knowledge, our work is the first theoretical treatment towards the degenerating aspect of multi-modal learning in neural networks.

In this work, we pursue a unified paradigm for multimodal pretraining to break the shackles of complex task/modality-specific customization. We propose OFA, a Task-Agnostic and Modality-Agnostic framework that supports Task Comprehensiveness. OFA unifies a diverse set of cross-modal and unimodal tasks, including image generation, visual grounding, image captioning, image classification, language modeling, etc., in a simple sequence-to-sequence learning framework. OFA follows the instruction-based learning in both pretraining and finetuning stages, requiring no extra task-specific layers for downstream tasks. In comparison with the recent state-of-the-art vision & language models that rely on extremely large cross-modal datasets, OFA is pretrained on only 20M publicly available image-text pairs. Despite its simplicity and relatively small-scale training data, OFA achieves new SOTAs in a series of cross-modal tasks while attaining highly competitive performances on uni-modal tasks. Our further analysis indicates that OFA can also effectively transfer to unseen tasks and unseen domains. Our code and models are publicly available at https://github.com/OFA-Sys/OFA.

Self-supervised learning (SSL) has been extensively explored in recent years. Particularly, generative SSL has seen emerging success in natural language processing and other fields, such as the wide adoption of BERT and GPT. Despite this, contrastive learning---which heavily relies on structural data augmentation and complicated training strategies---has been the dominant approach in graph SSL, while the progress of generative SSL on graphs, especially graph autoencoders (GAEs), has thus far not reached the potential as promised in other fields. In this paper, we identify and examine the issues that negatively impact the development of GAEs, including their reconstruction objective, training robustness, and error metric. We present a masked graph autoencoder GraphMAE (code is publicly available at https://github.com/THUDM/GraphMAE) that mitigates these issues for generative self-supervised graph learning. Instead of reconstructing structures, we propose to focus on feature reconstruction with both a masking strategy and scaled cosine error that benefit the robust training of GraphMAE. We conduct extensive experiments on 21 public datasets for three different graph learning tasks. The results manifest that GraphMAE---a simple graph autoencoder with our careful designs---can consistently generate outperformance over both contrastive and generative state-of-the-art baselines. This study provides an understanding of graph autoencoders and demonstrates the potential of generative self-supervised learning on graphs.

Academic Knowledge Services have substantially facilitated the development of human science and technology, providing a plenitude of useful research tools. However, many applications highly depend on ad-hoc models and expensive human labeling to understand professional contents, hindering deployments in real world. To create a unified backbone language model for various knowledge-intensive academic knowledge mining challenges, based on the world's largest public academic graph Open Academic Graph (OAG), we pre-train an academic language model, namely OAG-BERT, to integrate massive heterogeneous entity knowledge beyond scientific corpora. We develop novel pre-training strategies along with zero-shot inference techniques. OAG-BERT's superior performance on 9 knowledge-intensive academic tasks (including 2 demo applications) demonstrates its qualification to serve as a foundation for academic knowledge services. Its zero-shot capability also offers great potential to mitigate the need of costly annotations. OAG-BERT has been deployed to multiple real-world applications, such as reviewer recommendations for NSFC (National Nature Science Foundation of China) and paper tagging in the AMiner system. All codes and pre-trained models are available via the CogDL.

On-device machine learning enables the lightweight deployment of recommendation models in local clients, which reduces the burden of the cloud-based recommenders and simultaneously incorporates more real-time user features. Nevertheless, the cloud-based recommendation in the industry is still very important considering its powerful model capacity and the efficient candidate generation from the billion-scale item pool. Previous attempts to integrate the merits of both paradigms mainly resort to a sequential mechanism, which builds the on-device recommender on top of the cloud-based recommendation. However, such a design is inflexible when user interests dramatically change: the on-device model is stuck by the limited item cache while the cloud-based recommendation based on the large item pool do not respond without the new re-fresh feedback. To overcome this issue, we propose a meta controller to dynamically manage the collaboration between the on-device recommender and the cloud-based recommender, and introduce a novel efficient sample construction from the causal perspective to solve the dataset absence issue of meta controller. On the basis of the counterfactual samples and the extended training, extensive experiments in the industrial recommendation scenarios show the promise of meta controller in the device-cloud collaboration.

In this paper, we focus on unsupervised learning for Video Object Segmentation (VOS) which learns visual correspondence (i.e., the similarity between pixel-level features) from unlabeled videos. Previous methods are mainly based on the contrastive learning paradigm, which optimize either in image level or pixel level. Image-level optimization (e.g., the spatially pooled feature of ResNet) learns robust high-level semantics but is sub-optimal since the pixel-level features are optimized implicitly. By contrast, pixel-level optimization is more explicit, however, it is sensitive to the visual quality of training data and is not robust to object deformation. To complementarily perform these two levels of optimization in a unified framework, we propose the In-aNd-Out (INO) generative learning from a purely generative perspective with the help of naturally designed class tokens and patch tokens in Vision Transformer (ViT). Specifically, for image-level optimization, we force the out-view imagination from local to global views on class tokens, which helps capture high-level semantics, and we name it as out-generative learning. As to pixel-level optimization, we perform in-view masked image modeling on patch tokens, which recovers the corrupted parts of an image via inferring its fine-grained structure, and we term it as in-generative learning. To discover the temporal information better, we additionally force the inter-frame consistency from both feature and affinity matrix levels. Extensive experiments on DAVIS-2017 val and YouTube-VOS 2018 val show that our INO outperforms previous state-of-the-art methods by significant margins.

Large-scale pretrained AI models have shown state-of-the-art accuracy in a series of important applications. As the size of pretrained AI models grows dramatically each year in an effort to achieve higher accuracy, training such models requires massive computing and memory capabilities, which accelerates the convergence of AI and HPC. However, there are still gaps in deploying AI applications on HPC systems, which need application and system co-design based on specific hardware features.

In recent years, online short videos have become more popular, especially as an online advertising intermediary. To better understand their effects as advertisements, it is essential to analyze the causal relations of online short videos on consumer behaviors. Our study is based on fine-grained consumer behavior data from a world-leading e-commerce platform, i.e., Taobao.com. We first decompose the total causal effects into informative effects and persuasive effects following a common practice in the economic literature. Moreover, we extract the subjectivity scores of short videos through a dictionary-based subjectivity analysis model and evaluate the correlation between the subjectivity scores and each causal effect. The findings of this paper are as follows: First, both causal effects (i.e., informative and persuasive effects) are significant. Second, these effects have a strong correlation with the short videos' subjectivity scores. Third, the signs of these correlations vary with the prices of the products. Our results not only shed light on the research of how short videos exert influence on online consumers, but also give sellers advice on better video design and recommendation.

Recommendation system has been a widely studied task both in academia and industry. Previous works mainly focus on homogeneous recommendation and little progress has been made for heterogeneous recommender systems. However, heterogeneous recommendations, e.g., recommending different types of items including products, videos, celebrity shopping notes, among many others, are dominant nowadays. State-of-the-art methods are incapable of leveraging attributes from different types of items and thus suffer from data sparsity problems. And it is indeed quite challenging to represent items with different feature spaces jointly. To tackle this problem, we propose a kernel-based neural network, namely deep unified representation (or DURation) for heterogeneous recommendation, to jointly model unified representations of heterogeneous items while preserving their original feature space topology structures. Theoretically, we prove the representation ability of the proposed model. Besides, we conduct extensive experiments on the real-world datasets. Experimental results demonstrate that with the unified representation, our model achieves remarkable improvement (e.g., 4.1% ~34.9% lift by AUC score and 3.7% lift by online CTR) over existing state-of-the-art models.

Graph neural network-based recommendation systems are blossoming recently, and its core component is aggregation methods that determine neighbor embedding learning. Prior arts usually focus on how to aggregate information from the perspective of spatial structure information, but temporal information about neighbors is left insufficiently explored.

Sequential recommendation has become increasingly essential in various online services. It aims to model the dynamic preferences of users from their historical interactions and predict their next items. The accumulated user behavior records on real systems could be very long. This rich data brings opportunities to track actual interests of users. Prior efforts mainly focus on making recommendations based on relatively recent behaviors. However, the overall sequential data may not be effectively utilized, as early interactions might affect users' current choices. Also, it has become intolerable to scan the entire behavior sequence when performing inference for each user, since real-world system requires short response time. To bridge the gap, we propose a novel long sequential recommendation model, called Dynamic Memory-based Attention Network (DMAN). It segments the overall long behavior sequence into a series of sub-sequences, then trains the model and maintains a set of memory blocks to preserve long-term interests of users. To improve memory fidelity, DMAN dynamically abstracts each user's long-term interest into its own memory blocks by minimizing an auxiliary reconstruction loss. Based on the dynamic memory, the user's short-term and long-term interests can be explicitly extracted and combined for efficient joint recommendation. Empirical results over four benchmark datasets demonstrate the superiority of our model in capturing long-term dependency over various state-of-the-art sequential models.

Machine learning in the context of noise is a challenging but practical setting to plenty of real-world applications. Most of the previous approaches in this area focus on the pairwise relation (casual or correlational relationship) with noise, such as learning with noisy labels. However, the group noise, which is parasitic on the coarse-grained accurate relation with the fine-grained uncertainty, is also universal and has not been well investigated. The challenge under this setting is how to discover true pairwise connections concealed by the group relation with its fine-grained noise. To overcome this issue, we propose a novel Max-Matching method for learning with group noise. Specifically, it utilizes a matching mechanism to evaluate the relation confidence of each object w.r.t. the target, meanwhile considers the Non-IID characteristics among objects in the group. Only the most confident one of the objects is used to learn the model, so that the fine-grained noise is mostly dropped. The performance on a range of real-world datasets in the area of several learning paradigms demonstrates the effectiveness of Max-Matching.

Shuhuai Ren, Junyang Lin, Guangxiang Zhao, Rui Men, An Yang, Jingren Zhou, Xu Sun, Hongxia Yang. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Recommender system plays a crucial role in modern E-commerce platform. Due to the lack of historical interactions between users and items, cold-start recommendation is a challenging problem. In order to alleviate the cold-start issue, most existing methods introduce content and contextual information as the auxiliary information. Nevertheless, these methods assume the recommended items behave steadily over time, while in a typical E-commerce scenario, items generally have very different performances throughout their life period. In such a situation, it would be beneficial to consider the long-term return from the item perspective, which is usually ignored in conventional methods. Reinforcement learning (RL) naturally fits such a long-term optimization problem, in which the recommender could identify high potential items, proactively allocate more user impressions to boost their growth, therefore improve the multi-period cumulative gains. Inspired by this idea, we model the process as a Partially Observable and Controllable Markov Decision Process (POC-MDP), and propose an actor-critic RL framework (RL-LTV) to incorporate the item lifetime values (LTV) into the recommendation. In RL-LTV, the critic studies historical trajectories of items and predict the future LTV of fresh item, while the actor suggests a score-based policy which maximizes the future LTV expectation. Scores suggested by the actor are then combined with classical ranking scores in a dual-rank framework, therefore the recommendation is balanced with the LTV consideration. Our method outperforms the strong live baseline with a relative improvement of 8.67% and 18.03% on IPV and GMV of cold-start items, on one of the largest E-commerce platform.

Many existing neural architecture search (NAS) solutions rely on downstream training for architecture evaluation, which takes enormous computations. Considering that these computations bring a large carbon footprint, this paper aims to explore a green (namely environmental-friendly) NAS solution that evaluates architectures without training. Intuitively, gradients, induced by the architecture itself, directly decide the convergence and generalization results. It motivates us to propose the gradient kernel hypothesis: Gradients can be used as a coarse-grained proxy of downstream training to evaluate random-initialized networks. To support the hypothesis, we conduct a theoretical analysis and find a practical gradient kernel that has good correlations with training loss and validation performance. According to this hypothesis, we propose a new kernel based architecture search approach KNAS. Experiments show that KNAS achieves competitive results with orders of magnitude faster than “train-then-test” paradigms on image classification tasks. Furthermore, the extremely low search cost enables its wide applications. The searched network also outperforms strong baseline RoBERTA-large on two text classification tasks.

Existing reasoning tasks often have an important assumption that the input contents can be always accessed while reasoning, requiring unlimited storage resources and suffering from severe time delay on long sequences. To achieve efficient reasoning on long sequences with limited storage resources, memory augmented neural networks introduce a human-like write-read memory to compress and memorize the long input sequence in one pass, trying to answer subsequent queries only based on the memory. But they have two serious drawbacks: 1) they continually update the memory from current information and inevitably forget the early contents; 2) they do not distinguish what information is important and treat all contents equally. In this paper, we propose the Rehearsal Memory (RM) to enhance long-sequence memorization by self-supervised rehearsal with a history sampler. To alleviate the gradual forgetting of early information, we design self-supervised rehearsal training with recollection and familiarity tasks. Further, we design a history sampler to select informative fragments for rehearsal training, making the memory focus on the crucial information. We evaluate the performance of our rehearsal memory by the synthetic bAbI task and several downstream tasks, including text/video question answering and recommendation on long sequences.

Multimodal pretraining has demonstrated success in the downstream tasks of cross-modal representation learning. However, it is limited to the English data, and there is still a lack of large-scale dataset for multimodal pretraining in Chinese. In this work, we propose the largest dataset for pretraining in Chinese, which consists of over 1.9TB images and 292GB texts. The dataset has large coverage over domains, including encyclopedia, question answering, forum discussion, etc. Besides, we propose a method called M6, referring to Multi-Modality-to-Multi-Modality Multitask Mega-transformer, for unified pretraining on the data of single modality and multiple modalities. The model is pretrained with our proposed tasks, including text-to-text transfer, image-to-text transfer, as well as multi-modality-to-text transfer. The tasks endow the model with strong capability of understanding and generation. We scale the model to 10 billion parameters, and build the largest pretrained model in Chinese. Experimental results show that our proposed M6 outperforms the baseline in a number of downstream tasks concerning both single modality and multiple modalities, and the 10B-parameter pretrained model demonstrates strong potential in the setting of zero-shot learning.

With the rapid development of storage and computing power on mobile devices, it becomes critical and popular to deploy models on devices to save onerous communication latencies and to capture real-time features. While quite a lot of works have explored to facilitate on-device learning and inference, most of them focus on dealing with response delay or privacy protection. Little has been done to model the collaboration between the device and the cloud modeling and benefit both sides jointly. To bridge this gap, we are among the first attempts to study the Device-Cloud Collaborative Learning (DCCL) framework. Specifically, we propose a novel MetaPatch learning approach on the device side to efficiently achieve "thousands of people with thousands of models'' given a centralized cloud model. Then, with billions of updated personalized device models, we propose a "model-over-models'' distillation algorithm, namely MoMoDistill, to update the centralized cloud model. Our extensive experiments over a range of datasets with different settings demonstrate the effectiveness of such collaboration on both cloud and devices, especially its superiority to model long-tailed users.

Deep candidate generation (DCG) that narrows down the collection of relevant items from billions to hundreds via representation learning has become prevalent in industrial recommender systems. Standard approaches approximate maximum likelihood estimation (MLE) through sampling for better scalability and address the problem of DCG in a way similar to language modeling. However, live recommender systems face severe exposure bias and have a vocabulary several orders of magnitude larger than that of natural language, implying that MLE will preserve and even exacerbate the exposure bias in the long run in order to faithfully fit the observed samples. In this paper, we theoretically prove that a popular choice of contrastive loss is equivalent to reducing the exposure bias via inverse propensity weighting, which provides a new perspective for understanding the effectiveness of contrastive learning. Based on the theoretical discovery, we design CLRec, a contrastive learning method to improve DCG in terms of fairness, effectiveness and efficiency in recommender systems with extremely large candidate size. We further improve upon CLRec and propose Multi-CLRec, for accurate multi-intention aware bias reduction. Our methods have been successfully deployed in Taobao, where at least four-month online A/B tests and offline analyses demonstrate its substantial improvements, including a dramatic reduction in the Matthew effect.

Large-scale pre-trained language models have demonstrated strong capabilities of generating realistic texts. However, it remains challenging to control the generation results. Previous approaches such as prompting are far from sufficient, and lack of controllability limits the usage of language models. To tackle this challenge, we propose an innovative method, inverse prompting, to better control text generation. The core idea of inverse prompting is to use generated text to inversely predict the prompt during beam search, which enhances the relevance between the prompt and the generated text and thus improves controllability. Empirically, we pre-train a large-scale Chinese language model to perform a systematic study using human evaluation on the tasks of open-domain poem generation and open-domain long-form question answering. Results demonstrate that our proposed method substantially outperforms the baselines and that our generation quality is close to human performance on some of the tasks.

Text-to-Image generation in the general domain has long been an open problem, which requires both a powerful generative model and cross-modal understanding. We propose CogView, a 4-billion-parameter Transformer with VQ-VAE tokenizer to advance this problem. We also demonstrate the finetuning strategies for various downstream tasks, e.g. style learning, super-resolution, text-image ranking and fashion design, and methods to stabilize pretraining, e.g. eliminating NaN losses. CogView achieves the state-of-the-art FID on the blurred MS COCO dataset, outperforming previous GAN-based models and a recent similar work DALL-E.

Conditional image synthesis aims to create an image according to some multi-modal guidance in the forms of textual descriptions, reference images, and image blocks to preserve, as well as their combinations. In this paper, instead of investigating these control signals separately, we propose a new two-stage architecture, UFC-BERT, to unify any number of multi-modal controls. In UFC-BERT, both the diverse control signals and the synthesized image are uniformly represented as a sequence of discrete tokens to be processed by Transformer. Different from existing two-stage autoregressive approaches such as DALL-E and VQGAN, UFC-BERT adopts non-autoregressive generation (NAR) at the second stage to enhance the holistic consistency of the synthesized image, to support preserving specified image blocks, and to improve the synthesis speed. Further, we design a progressive algorithm that iteratively improves the non-autoregressively generated image, with the help of two estimators developed for evaluating the compliance with the controls and evaluating the fidelity of the synthesized image, respectively. Extensive experiments on a newly collected large-scale clothing dataset M2C-Fashion and a facial dataset Multi-Modal CelebA-HQ verify that UFC-BERT can synthesize high-fidelity images that comply with flexible multi-modal controls.

Recent methods in sequential recommendation focus on learning an overall embedding vector from a user's behavior sequence for the next-item recommendation. However, from empirical analysis, we discovered that a user's behavior sequence often contains multiple conceptually distinct items, while a unified embedding vector is primarily affected by one's most recent frequent actions. Thus, it may fail to infer the next preferred item if conceptually similar items are not dominant in recent interactions. To this end, an alternative solution is to represent each user with multiple embedding vectors encoding different aspects of the user's intentions. Nevertheless, recent work on multi-interest embedding usually considers a small number of concepts discovered via clustering, which may not be comparable to the large pool of item categories in real systems. It is a non-trivial task to effectively model a large number of diverse conceptual prototypes, as items are often not conceptually well clustered in fine granularity. Besides, an individual usually interacts with only a sparse set of concepts. In light of this, we propose a novel Sparse Interest NEtwork (SINE) for sequential recommendation. Our sparse-interest module can adaptively infer a sparse set of concepts for each user from the large concept pool and output multiple embeddings accordingly. Given multiple interest embeddings, we develop an interest aggregation module to actively predict the user's current intention and then use it to explicitly model multiple interests for next-item prediction. Empirical results on several public benchmark datasets and one large-scale industrial dataset demonstrate that SINE can achieve substantial improvement over state-of-the-art methods.

In this paper, we identify and study an important problem of gradient item retrieval. We define the problem as retrieving a sequence of items with a gradual change on a certain attribute, given a reference item and a modification text. For example, after a customer saw a white dress, she/he wants to buy a similar one but more floral on it. The extent of ”more floral” is subjective, thus prompting one floral dress is hard to satisfy the customer’s needs. A better way is to present a sequence of products with increasingly floral attributes based on the white dress, and allow the customer to select the most satisfactory one from the sequence. Existing item retrieval methods mainly focus on whether the target items appear at the top of the retrieved sequence, but ignore the demand for retrieving a sequence of products with gradual change on a certain attribute. To deal with this problem, we propose a weakly-supervised method that can learn a disentangled item representation from user-item interaction data and ground the semantic meaning of attributes to dimensions of the item representation. Our method takes a reference item and a modification as a query. During inference, we start from the reference item and ”walk” along the direction of the modification in the item representation space to retrieve a sequence of items in a gradient manner. We demonstrate our proposed method can achieve disentanglement through weak supervision. Besides, we empirically show that an item sequence retrieved by our method is gradually changed on an indicated attribute and, in the item retrieval task, our method outperforms existing approaches on three different datasets.

Self-attention has become increasingly popular in a variety of sequence modeling tasks from natural language processing to recommendation, due to its effectiveness. However, self-attention suffers from quadratic computational and memory complexities, prohibiting its applications on long sequences. Existing approaches that address this issue mainly rely on a sparse attention context, either using a local window, or a permuted bucket obtained by locality-sensitive hashing (LSH) or sorting, while crucial information may be lost. Inspired by the idea of vector quantization that uses cluster centroids to approximate items, we propose LISA (LInear-time Self Attention), which enjoys both the effectiveness of vanilla self-attention and the efficiency of sparse attention. LISA scales linearly with the sequence length, while enabling full contextual attention via computing differentiable histograms of codeword distributions. Meanwhile, unlike some efficient attention methods, our method poses no restriction on casual masking or sequence length. We evaluate our method on four real-world datasets for sequential recommendation. The results show that LISA outperforms the state-of-the-art efficient attention methods in both performance and speed; and it is up to 57x faster and 78x more memory efficient than vanilla self-attention.

Many real-world applications involve data from multiple modalities and thus exhibit the view heterogeneity. For example, user modeling on social media might leverage both the topology of the underlying social network and the content of the users’ posts; in the medical domain, multiple views could be X-ray images taken at different poses. To date, various techniques have been proposed to achieve promising results, such as canonical correlation analysis based methods, etc. In the meanwhile, it is critical for decision-makers to be able to understand the prediction results from these methods. For example, given the diagnostic result that a model provided based on the X-ray images of a patient at different poses, the doctor needs to know why the model made such a prediction. However, state-of-the-art techniques usually suffer from the inability to utilize the complementary information of each view and to explain the predictions in an interpretable manner.

In real-world problems, heterogeneous entities are often related to each other through multiple interactions, forming a Heterogeneous Interaction Graph (HIG). While modeling HIGs to deal with fundamental tasks, graph neural networks present an attractive opportunity that can make full use of the heterogeneity and rich semantic information by aggregating and propagating information from different types of neighborhoods. However, learning on such complex graphs, often with millions or billions of nodes, edges, and various attributes, could suffer from expensive time cost and high memory consumption. In this article, we attempt to accelerate representation learning on large-scale HIGs by adopting the importance sampling of heterogeneous neighborhoods in a batch-wise manner, which naturally fits with most batch-based optimizations. Distinct from traditional homogeneous strategies neglecting semantic types of nodes and edges, to handle the rich heterogeneous semantics within HIGs, we devise both type-dependent and type-fusion samplers where the former respectively samples neighborhoods of each type and the latter jointly samples from candidates of all types. Furthermore, to overcome the imbalance between the down-sampled and the original information, we respectively propose heterogeneous estimators including the self-normalized and the adaptive estimators to improve the robustness of our sampling strategies. Finally, we evaluate the performance of our models for node classification and link prediction on five real-world datasets, respectively. The empirical results demonstrate that our approach performs significantly better than other state-of-the-art alternatives, and is able to reduce the number of edges in computation by up to 93%, the memory cost by up to 92% and the time cost by up to 86%.

Network embedding aims to automatically learn the node representations in networks. The basic idea of network embedding is to first construct a network to describe the neighborhood context for each node, and then learn the node representations by designing an objective function to preserve certain properties of the constructed context network. The vast majority of the existing methods, explicitly or implicitly, follow a pointwise design principle. That is, the objective can be decomposed into the summation of the certain goodness function over each individual edge of the context network. In this paper, we propose to go beyond such pointwise approaches, and introduce the ranking-oriented design principle for network embedding. The key idea is to decompose the overall objective function into the summation of a goodness function over a set of edges to collectively preserve their relative rankings on the context network. We instantiate the ranking-oriented design principle by two new network embedding algorithms, including a pairwise network embedding method PaWine which optimizes the relative weights of edge pairs, and a listwise method LiWine which optimizes the relative weights of edge lists. Both proposed algorithms bear a linear time complexity, making themselves scalable to large networks. We conduct extensive experimental evaluations on five real datasets with a variety of downstream learning tasks, which demonstrate that the proposed approaches consistently outperform the existing methods.

Granger causal modeling is an emerging topic that can uncover Granger causal relationship behind multivariate time series data. In many real-world systems, it is common to encounter a large amount of multivariate time series data collected from different individuals with sharing commonalities. However, there are ongoing concerns regarding Granger causality's applicability in such large scale complex scenarios, presenting both challenges and opportunities for Granger causal structure reconstruction. Existing methods usually train a distinct model for each individual, suffering from inefficiency and over-fitting issues. To bridge this gap, we propose an Inductive GRanger cAusal modeling (InGRA) framework for inductive Granger causality learning and common causal structure detection on multivariate time series, which exploits the shared commonalities underlying the different individuals. In particular, we train one global model for individuals with different Granger causal structures through a novel attention mechanism, called prototypical Granger causal attention. The model can detect common causal structures for different individuals and infer Granger causal structures for newly arrived individuals. Extensive experiments, as well as an online A/B test on an E-commercial advertising platform, demonstrate the superior performances of InGRa.

Nowadays, both online shopping and video sharing have grown exponentially. Although internet celebrities in videos are ideal exhibition for fashion corporations to sell their products, audiences do not always know where to buy fashion products in videos, which is a cross-domain problem called video-to-shop. In this paper, we propose a novel deep neural network, called Detect, Pick, and Retrieval Network (DPRNet), to break the gap between fashion products from videos and audiences. For the video side, we have modified the traditional object detector, which automatically picks out the best object proposals for every commodity in videos without duplication, to promote the performance of the video-to-shop task. For the fashion retrieval side, a simple but effective multi-task loss network obtains new state-of-the-art results on DeepFashion. Extensive experiments conducted on a new large-scale cross-domain video-to-shop dataset shows that DPRNet is efficient and outperforms the state-of-the-art methods on video-to-shop task.

Recently, neural networks have been widely used in e-commerce recommender systems, owing to the rapid development of deep learning. We formalize the recommender system as a sequential recommendation problem, intending to predict the next items that the user might be interacted with. Recent works usually give an overall embedding from a user's behavior sequence. However, a unified user embedding cannot reflect the user's multiple interests during a period. In this paper, we propose a novel controllable multi-interest framework for the sequential recommendation, called ComiRec. Our multi-interest module captures multiple interests from user behavior sequences, which can be exploited for retrieving candidate items from the large-scale item pool. These items are then fed into an aggregation module to obtain the overall recommendation. The aggregation module leverages a controllable factor to balance the recommendation accuracy and diversity. We conduct experiments for the sequential recommendation on two real-world datasets, Amazon and Taobao. Experimental results demonstrate that our framework achieves significant improvements over state-of-the-art models. Our framework has also been successfully deployed on the offline Alibaba distributed cloud platform.

Nowadays graph has become a general and powerful representation to describe the rich relationships among different kinds of entities via the underlying patterns encoded in its structure. The knowledge (more generally) accumulated in graph is expected to be able to cross populations from one to another and the past to future. However the data collection process of graph generation is full of known or unknown sample selection biases, leading to spurious correlations among entities, especially in the non-stationary and heterogeneous environments. In this paper, we target the problem of learning stable graphs from multiple environments with selection bias. We purpose a Stable Graph Learning (SGL) framework to learn a graph that can capture general relational patterns which are irrelevant with the selection bias in an unsupervised way. Extensive experimental results from both simulation and real data demonstrate that our method could significantly benefit the generalization capacity of graph structure.

To learn a sequential recommender, the existing methods typically adopt the sequence-to-item (seq2item) training strategy, which supervises a sequence model with a user's next behavior as the label and the user's past behaviors as the input. The seq2item strategy, however, is myopic and usually produces non-diverse recommendation lists. In this paper, we study the problem of mining extra signals for supervision by looking at the longer-term future. There exist two challenges: i) reconstructing a future sequence containing many behaviors is exponentially harder than reconstructing a single next behavior, which can lead to difficulty in convergence, and ii) the sequence of all future behaviors can involve many intentions, not all of which may be predictable from the sequence of earlier behaviors. To address these challenges, we propose a sequence-to-sequence (seq2seq) training strategy based on latent self-supervision and disentanglement. Specifically, we perform self-supervision in the latent space, i.e., reconstructing the representation of the future sequence as a whole, instead of reconstructing the items in the future sequence individually. We also disentangle the intentions behind any given sequence of behaviors and construct seq2seq training samples using only pairs of sub-sequences that involve a shared intention. Results on real-world benchmarks and synthetic data demonstrate the improvement brought by seq2seq training.

Graph representation learning has emerged as a powerful technique for addressing real-world problems. Various downstream graph learning tasks have benefited from its recent developments, such as node classification, similarity search, and graph classification. However, prior arts on graph representation learning focus on domain specific problems and train a dedicated model for each graph dataset, which is usually non-transferable to out-of-domain data. Inspired by the recent advances in pre-training from natural language processing and computer vision, we design Graph Contrastive Coding (GCC) --- a self-supervised graph neural network pre-training framework --- to capture the universal network topological properties across multiple networks. We design GCC's pre-training task as subgraph instance discrimination in and across networks and leverage contrastive learning to empower graph neural networks to learn the intrinsic and transferable structural representations. We conduct extensive experiments on three graph learning tasks and ten graph datasets. The results show that GCC pre-trained on a collection of diverse datasets can achieve competitive or better performance to its task-specific and trained-from-scratch counterparts. This suggests that the pre-training and fine-tuning paradigm presents great potential for graph representation learning.

Graph representation learning has been extensively studied in recent years, in which sampling is a critical point. Prior arts usually focus on sampling positive node pairs, while the strategy for negative sampling is left insufficiently explored. To bridge the gap, we systematically analyze the role of negative sampling from the perspectives of both objective and risk, theoretically demonstrating that negative sampling is as important as positive sampling in determining the optimization objective and the resulted variance. To the best of our knowledge, we are the first to derive the theory and quantify that a nice negative sampling distribution is pn(u|v) ∝ pd(u|v)α, 0 < α < 1. With the guidance of the theory, we propose MCNS, approximating the positive distribution with self-contrast approximation and accelerating negative sampling by Metropolis-Hastings. We evaluate our method on 5 datasets that cover extensive downstream graph learning tasks, including link prediction, node classification and recommendation, on a total of 19 experimental settings. These relatively comprehensive experimental results demonstrate its robustness and superiorities.

In e-commerce, consumer-generated videos, which in general deliver consumers' individual preferences for the different aspects of certain products, are massive in volume. To recommend these videos to potential consumers more effectively, diverse and catchy video titles are critical. However, consumer-generated videos seldom accompany appropriate titles. To bridge this gap, we integrate comprehensive sources of information, including the content of consumer-generated videos, the narrative comment sentences supplied by consumers, and the product attributes, in an end-to-end modeling framework. Although automatic video titling is very useful and demanding, it is much less addressed than video captioning. The latter focuses on generating sentences that describe videos as a whole while our task requires the product-aware multi-grained video analysis. To tackle this issue, the proposed method consists of two processes, i.e., granular-level interaction modeling and abstraction-level story-line summarization. Specifically, the granular-level interaction modeling first utilizes temporal-spatial landmark cues, descriptive words, and abstractive attributes to builds three individual graphs and recognizes the intra-actions in each graph through Graph Neural Networks (GNN). Then the global-local aggregation module is proposed to model inter-actions across graphs and aggregate heterogeneous graphs into a holistic graph representation. The abstraction-level story-line summarization further considers both frame-level video features and the holistic graph to utilize the interactions between products and backgrounds, and generate the story-line topic of the video. We collect a large-scale dataset accordingly from real-world data in Taobao, a world-leading e-commerce platform, and will make the desensitized version publicly available to nourish further development of the research community. Relatively extensive experiments on various datasets demonstrate the efficacy of the proposed method.

In this paper, we propose to investigate the problem of out-of-domain visio-linguistic pretraining, where the pretraining data distribution differs from that of downstream data on which the pretrained model will be fine-tuned. Existing methods for this problem are purely likelihood-based, leading to the spurious correlations and hurt the generalization ability when transferred to out-of-domain downstream tasks. By spurious correlation, we mean that the conditional probability of one token (object or word) given another one can be high (due to the dataset biases) without robust (causal) relationships between them. To mitigate such dataset biases, we propose a Deconfounded Visio-Linguistic Bert framework, abbreviated as DeVLBert, to perform intervention-based learning. We borrow the idea of the backdoor adjustment from the research field of causality and propose several neural-network based architectures for Bert-style out-of-domain pretraining. The quantitative results on three downstream tasks, Image Retrieval (IR), Zero-shot IR, and Visual Question Answering, show the effectiveness of DeVLBert by boosting generalization ability.

In e-commerce, a growing number of user-generated videos are used for product promotion. How to generate video descriptions that narrate the user-preferred product characteristics depicted in the video is vital for successful promoting. Traditional video captioning methods, which focus on routinely describing what exists and happens in a video, are not amenable for product-oriented video captioning. To address this problem, we propose a product-oriented video captioner framework, abbreviated as Poet. Poet firstly represents the videos as product-oriented spatial-temporal graphs. Then, based on the aspects of the video-associated product, we perform knowledge-enhanced spatial-temporal inference on those graphs for capturing the dynamic change of fine-grained product-part characteristics. The knowledge leveraging module in Poet differs from the traditional design by performing knowledge filtering and dynamic memory modeling. We show that Poet achieves consistent performance improvement over previous methods concerning generation quality, product aspects capturing, and lexical diversity. Experiments are performed on two product-oriented video captioning datasets, buyer-generated fashion video dataset (BFVD) and fan-generated fashion video dataset (FFVD), collected from Mobile Taobao. We will release the desensitized datasets to promote further investigations on both video captioning and general video analysis problems.

BERTs are incapable of processing long texts due to its quadratically increasing memory and time consumption. The straightforward thoughts to address this problem, such as slicing the text by a sliding window or simplifying transformers, suffer from insufficient long-range attentions or need customized CUDA kernels. The limited text length of BERT reminds us the limited capacity (5∼ 9 chunks) of the working memory of humans – then how do human beings Cognize Long TeXts? Founded on the cognitive theory stemming from Baddeley, our CogLTX framework identifies key sentences by training a judge model, concatenates them for reasoning and enables multi-step reasoning via rehearsal and decay. Since relevance annotations are usually unavailable, we propose to use treatment experiments to create supervision. As a general algorithm, CogLTX outperforms or gets comparable results to SOTA models on NewsQA, HotpotQA, multi-class and multi-label long-text classification tasks with memory overheads independent of the text length.

Estimating counterfactual outcome of different treatments from observational data is an important problem to assist decision making in a variety of fields. Among the various forms of treatment specification, bundle treatment has been widely adopted in many scenarios, such as recommendation systems and online marketing. The bundle treatment usually can be abstracted as a high dimensional binary vector, which makes it more challenging for researchers to remove the confounding bias in observational data. In this work, we assume the existence of low dimensional latent structure underlying bundle treatment. Via the learned latent representations of treatments, we propose a novel variational sample re-weighting (VSR) method to eliminate confounding bias by decorrelating the treatments and confounders. Finally, we conduct extensive experiments to demonstrate that the predictive model trained on this re-weighted dataset can achieve more accurate counterfactual outcome prediction.

Network embedding effectively transforms complex network data into a low-dimensional vector space and has shown great performance in many real-world scenarios, such as link prediction, node classification, and similarity search. A plethora of methods have been proposed to learn node representations and achieve encouraging results. Nevertheless, little attention has been paid on the embedding technique for bipartite attributed networks, which is a typical data structure for modeling nodes from two distinct partitions.

The recent success of deep graph embedding innovates the graphical information characterization methodologies. However, in real-world applications, such a method still struggles with the challenges of heterogeneity, scalability, and multiplex. To address these challenges, in this study, we propose a novel solution, Genetic hEterogeneous gRaph eMbedding (GERM), which enables flexible and efficient task-driven vertex embedding in a complex heterogeneous graph. Unlike prior efforts for this track of studies, we employ a task-oriented genetic activation strategy to efficiently generate the “Edge Type Activated Vector” (ETAV) over the edge types in the graph. The generated ETAV can not only reduce the incompatible noise and navigate the heterogeneous graph random walk at the graph-schema level, but also activate an optimized subgraph for efficient representation learning. By revealing the correlation between the graph structure and task information, the model interpretability can be enhanced as well. Meanwhile, an activated heterogeneous skip-gram framework is proposed to encapsulate both topological and task-specific information of a given heterogeneous graph. Through extensive experiments on both scholarly and e-commerce datasets, we demonstrate the efficacy and scalability of the proposed methods via various search/recommendation tasks. GERM can significantly reduces the running time and remove expert-intervention without sacrificing the performance (or even modestly improve) by comparing with baselines.

Recommender systems in industry generally include two stages: recall and ranking. Recall refers to efficiently identify hundreds of candidate items that user may interest in from a large volume of item corpus, while the latter aims to output a precise ranking list using complex ranking models. Recently, graph representation learning has attracted much attention in supporting high quality candidate search at scale. Despite its effectiveness in learning embedding vectors for objects in the user-item interaction network, the computational costs to infer users’ preferences in continuous embedding space are tremendous. In this work, we investigate the problem of hashing with graph neural networks (GNNs) for high quality retrieval, and propose a simple yet effective discrete representation learning framework to jointly learn continuous and discrete codes. Specifically, a deep hashing with GNNs (HashGNN) is presented, which consists of two components, a GNN encoder for learning node representations, and a hash layer for encoding representations to hash codes. The whole architecture is trained end-to-end by jointly optimizing two losses, i.e., reconstruction loss from reconstructing observed links, and ranking loss from preserving the relative ordering of hash codes. A novel discrete optimization strategy based on straight through estimator (STE) with guidance is proposed. The principal idea is to avoid gradient magnification in back-propagation of STE with continuous embedding guidance, in which we begin from learning an easier network that mimic the continuous embedding and let it evolve during the training until it finally goes back to STE. Comprehensive experiments over three publicly available and one real-world Alibaba company datasets demonstrate that our model not only can achieve comparable performance compared with its continuous counterpart but also runs multiple times faster during inference.

This paper introduces how to infer trust relationships from billion-scale networked data to benefit Alibaba E-Commerce business. To effectively leverage the network correlations between labeled and unlabeled relationships to predict trust relationships, we formalize trust into multiple types and propose a graphical model to incorporate type-based dyadic and triadic correlations, namely eTrust. We also present a fast learning algorithm in order to handle billion-scale networks. Systematically, we evaluate the proposed methods on four different genres of datasets with labeled trust relationships: Alibaba, Epinions, Ciao, and Advogato. Experimental results show that the proposed methods achieve significantly better performance than several comparison methods (+1.7-32.3% by accuracy; p <; <; 0:01, with t-test). Most importantly, when handling the real large networked data with over 1,200,000,000 edges (Ali-large), our method achieves 2,000× speedup to infer trust relationships, comparing with the traditional graph learning algorithms. Finally, we have applied the inferred trust relationships to Alibaba E-commerce platform: Taobao, and achieved 2.75 percent improvement on gross merchandise volume (GMV).

We propose a new CogQA framework for multi-hop reading comprehension question answering in web-scale documents. Founded on the dual process theory in cognitive science, the framework gradually builds a cognitive graph in an iterative process by coordinating an implicit extraction module (System 1) and an explicit reasoning module (System 2). While giving accurate answers, our framework further provides explainable reasoning paths. Specifically, our implementation based on BERT and graph neural network efficiently handles millions of documents for multi-hop reasoning questions in the HotpotQA fullwiki dataset, achieving a winning joint F1 score of 34.9 on the leaderboard, compared to 23.1 of the best competitor.

Low-dimensional embeddings of knowledge graphs and behavior graphs have proved remarkably powerful in varieties of tasks, from predicting unobserved edges between entities to content recommendation. The two types of graphs can contain distinct and complementary information for the same entities/nodes. However, previous works focus either on knowledge graph embedding or behavior graph embedding while few works consider both in a unified way. Here we present BEM, a Bayesian framework that incorporates the information from knowledge graphs and behavior graphs. To be more specific, BEM takes as prior the pre-trained embeddings from the knowledge graph, and integrates them with the pre-trained embeddings from the behavior graphs via a Bayesian generative model. BEM is able to mutually refine the embeddings from both sides while preserving their own topological structures. To show the superiority of our method, we conduct a range of experiments on three benchmark datasets: node classification, link prediction, triplet classification on two small datasets related to Freebase, and item recommendation on a large-scale e-commerce dataset.

Qibin Chen, Junyang Lin, Yichang Zhang, Ming Ding, Yukuo Cen, Hongxia Yang, Jie Tang. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Recommender systems on E-Commerce platforms track users' online behaviors and recommend relevant items according to each user’s interests and needs. Bipartite graphs that capture both user/item feature and use-item interactions have been demonstrated to be highly effective for this purpose. Recently, graph neural network (GNN) has been successfully applied in representation of bipartite graphs in industrial recommender systems. Providing individualized recommendation on a dynamic platform with billions of users is extremely challenging. A key observation is that the users of an online E-Commerce platform can be naturally clustered into a set of communities. We propose to cluster the users into a set of communities and make recommendations based on the information of the users in the community collectively. More specifically, embeddings are assigned to the communities and the user embedding is decomposed into two parts, each of which captures the community-level generalizations and individualized preferences respectively. The community embedding can be considered as an enhancement to the GNN methods that are inherently flat and do not learn hierarchical representations of graphs. The performance of the proposed algorithm is demonstrated on a public dataset and a world-leading E-Commerce company dataset.

Analyzing large-scale evolving graphs are crucial for understanding the dynamic and evolutionary nature of social networks. Most existing works focus on discovering repeated and consistent temporal patterns, however, such patterns cannot fully explain the complexity observed in dynamic networks. For example, in recommendation scenarios, users sometimes purchase products on a whim during a window shopping.Thus, in this paper, we design and implement a novel framework called BurstGraph which can capture both recurrent and consistent patterns, and especially unexpected bursty network changes. The performance of the proposed algorithm is demonstrated on both a simulated dataset and a world-leading E-Commerce company dataset, showing that they are able to discriminate recurrent events from extremely bursty events in terms of action propensity.

Network embedding (or graph embedding) has been widely used in many real-world applications. However, existing methods mainly focus on networks with single-typed nodes/edges and cannot scale well to handle large networks. Many real-world networks consist of billions of nodes and edges of multiple types, and each node is associated with different attributes. In this paper, we formalize the problem of embedding learning for the Attributed Multiplex Heterogeneous Network and propose a unified framework to address this problem. The framework supports both transductive and inductive learning. We also give the theoretical analysis of the proposed framework, showing its connection with previous works and proving its better expressiveness. We conduct systematical evaluations for the proposed framework on four different genres of challenging datasets: Amazon, YouTube, Twitter, and Alibaba. Experimental results demonstrate that with the learned embeddings from the proposed framework, we can achieve statistically significant improvements (e.g., 5.99-28.23% lift by F1 scores; p<<0.01, t-test) over previous state-of-the-art methods for link prediction. The framework has also been successfully deployed on the recommendation system of a worldwide leading e-commerce company, Alibaba Group. Results of the offline A/B tests on product recommendation further confirm the effectiveness and efficiency of the framework in practice.

Quality product descriptions are critical for providing competitive customer experience in an E-commerce platform. An accurate and attractive description not only helps customers make an informed decision but also improves the likelihood of purchase. However, crafting a successful product description is tedious and highly time-consuming. Due to its importance, automating the product description generation has attracted considerable interest from both research and industrial communities. Existing methods mainly use templates or statistical methods, and their performance could be rather limited. In this paper, we explore a new way to generate personalized product descriptions by combining the power of neural networks and knowledge base. Specifically, we propose a KnOwledge Based pErsonalized (or KOBE) product description generation model in the context of E-commerce.

Cold-start problems are long-standing challenges for practical recommendations. Most existing recommendation algorithms rely on extensive observed data and are brittle to recommendation scenarios with few interactions. This paper addresses such problems usingfew-shot learning andmeta learning. Our approach is based on the insight that having a good generalization from a few examples relies on both a generic model initialization and an effective strategy for adapting this model to newly arising tasks. To accomplish this, we combine the scenario-specific learning with a model-agnostic sequential meta-learning and unify them into an integrated end-to-end framework, namely S cenario-specific S equential Meta learner (or s^2Meta). By doing so, ourmeta-learner produces a generic initial model through aggregating contextual information from a variety of prediction tasks while effectively adapting to specific tasks by leveraging learning-to-learn knowledge. Extensive experiments on various real-world datasets demonstrate that our proposed model can achieve significant gains over the state-of-the-arts for cold-start problems in online recommendation. Deployment is at the Guess You Like session, the front page of the Mobile Taobao; and the illustration video can also be watched from the link\footnote\urlhttps://youtu.be/TNHLZqWnQwc .

Arguably, every entity in this universe is networked in one wayr another. With the prevalence of network data collected, such as social media and biological networks, learning from networks has become an essential task in many applications. It is well recognized that network data is intricate and large-scale, and analytic tasks on network data become more and more sophisticated. In this tutorial, we systematically review the area of learning from networks, including algorithms, theoretical analysis, and illustrative applications. Starting with a quick recollection of the exciting history of the area, we formulate the core technical problems. Then, we introduce the fundamental approaches, that is, the feature selection based approaches and the network embedding based approaches. Next, we extend our discussion to attributed networks, which are popular in practice. Last, we cover the latest hot topic, graph neural based approaches. For each group of approaches, we also survey the associated theoretical analysis and real-world application examples. Our tutorial also inspires a series of open problems and challenges that may lead to future breakthroughs. The authors are productive and seasoned researchers active in this area who represent a nice combination of academia and industry.

Networks have been widely used as the data structure for abstracting real-world systems as well as organizing the relations among entities. Network embedding models are powerful tools in mapping nodes in a network into continuous vector-space representations in order to facilitate subsequent tasks such as classification and link prediction. Existing network embedding models comprehensively integrate all information of each node, such as links and attributes, towards a single embedding vector to represent the node's general role in the network. However, a real-world entity could be multifaceted, where it connects to different neighborhoods due to different motives or self-characteristics that are not necessarily correlated. For example, in a movie recommender system, a user may love comedies or horror movies simultaneously, but it is not likely that these two types of movies are mutually close in the embedding space, nor the user embedding vector could be sufficiently close to them at the same time. In this paper, we propose a polysemous embedding approach for modeling multiple facets of nodes, as motivated by the phenomenon of word polysemy in language modeling. Each facet of a node is mapped to an embedding vector, while we also maintain an association degree between each pair of node and facet. The proposed method is adaptive to various existing embedding models, without significantly complicating the optimization process. We also discuss how to engage embedding vectors of different facets for inference tasks including classification and link prediction. Experiments on real-world datasets help comprehensively evaluate the performance of the proposed method.

An increasing number of machine learning tasks require dealing with large graph datasets, which capture rich and complex relation- ship among potentially billions of elements. Graph Neural Network (GNN) becomes an effective way to address the graph learning problem by converting the graph data into a low dimensional space while keeping both the structural and property information to the maximum extent and constructing a neural network for training and referencing. However, it is challenging to provide an efficient graph storage and computation capabilities to facilitate GNN training and enable development of new GNN algorithms. In this paper, we present a comprehensive graph neural network system, namely AliGraph, which consists of distributed graph storage, optimized sampling operators and runtime to efficiently support not only existing popular GNNs but also a series of in-house developed ones for different scenarios. The system is currently deployed at Alibaba to support a variety of business scenarios, including product recommendation and personalized search at Alibaba's E-Commerce platform. By conducting extensive experiments on a real-world dataset with 492.90 million vertices, 6.82 billion edges and rich attributes, Ali- Graph performs an order of magnitude faster in terms of graph building (5 minutes vs hours reported from the state-of-the-art PowerGraph platform). At training, AliGraph runs 40%-50% faster with the novel caching strategy and demonstrates around 12 times speed up with the improved runtime. In addition, our in-house developed GNN models all showcase their statistically significant superiorities in terms of both effectiveness and efficiency (e.g., 4.12% 17.19% lift by F1 scores).

User behavior data in recommender systems are driven by the complex interactions of many latent factors behind the users’ decision making processes. The factors are highly entangled, and may range from high-level ones that govern user intentions, to low-level ones that characterize a user’s preference when executing an intention. Learning representations that uncover and disentangle these latent factors can bring enhanced robustness, interpretability, and controllability. However, learning such disentangled representations from user behavior is challenging, and remains largely neglected by the existing literature. In this paper, we present the MACRo-mIcro Disentangled Variational Auto-Encoder (MacridVAE) for learning disentangled representations from user behavior. Our approach achieves macro disentanglement by inferring the high-level concepts associated with user intentions (e.g., to buy a shirt or a cellphone), while capturing the preference of a user regarding the different concepts separately. A micro-disentanglement regularizer, stemming from an information-theoretic interpretation of VAEs, then forces each dimension of the representations to independently reflect an isolated low-level factor (e.g., the size or the color of a shirt). Empirical results show that our approach can achieve substantial improvement over the state-of-the-art baselines. We further demonstrate that the learned representations are interpretable and controllable, which can potentially lead to a new paradigm for recommendation where users are given fine-grained control over targeted aspects of the recommendation lists.

An increasing number of machine learning tasks require dealing with large graph datasets, which capture rich and complex relationship among potentially billions of elements. Graph Neural Network (GNN) becomes an effective way to address the graph learning problem by converting the graph data into a low dimensional space while keeping both the structural and property information to the maximum extent and constructing a neural network for training and referencing. However, it is challenging to provide an efficient graph storage and computation capabilities to facilitate GNN training and enable development of new GNN algorithms. In this paper, we present a comprehensive graph neural network system, namely AliGraph , which consists of distributed graph storage, optimized sampling operators and runtime to efficiently support not only existing popular GNNs but also a series of in-house developed ones for different scenarios. The system is currently deployed at Alibaba to support a variety of business scenarios, including product recommendation and personalized search at Alibaba's E-Commerce platform. By conducting extensive experiments on a real-world dataset with 492.90 million vertices, 6.82 billion edges and rich attributes, AliGraph performs an order of magnitude faster in terms of graph building (5 minutes vs hours reported from the state-of-the-art PowerGraph platform). At training, AliGraph runs 40%-50% faster with the novel caching strategy and demonstrates around 12 times speed up with the improved runtime. In addition, our in-house developed GNN models all showcase their statistically significant superiorities in terms of both effectiveness and efficiency (e.g., 4.12%--17.19% lift by F1 scores).

Attributed network embedding focuses on learning low-dimensional latent representations of nodes which can well preserve the original topological and node attributed proximity at the same time. Existing works usually assume that nodes with similar topology or similar attributes should also be close in the embedding space. This assumption ignores the phenomenon of partial correlation between network topological and node attributed similarities i.e. nodes with similar topology may be dissimilar in their attributes and vice versa. Partial correlation between the two information sources should be considered especially when there exist fraudulent edges (i.e., information from one source is vague) or unbalanced data distributions (i.e, topology structure similarity and node attribute similarity have different distributions). However, it is very challenging to consider the partial correlation between topology and attributes due to the heterogeneity of these two information sources. In this paper, we take partial correlation between topology and attributes into account and propose the Personalized Relation Ranking Embedding (PRRE) method for attributed networks which is capable of exploiting the partial correlation between node topology and attributes. The proposed PRRE model utilizes two thresholds to define different node relations and employs the Expectation-Maximization (EM) algorithm to learn these thresholds as well as other embedding parameters. Extensive experiments results on multiple real-world datasets show that the proposed PRRE model significantly outperforms the state-of-the-art methods in terms of various evaluation metrics.

We study the important problem of user alignment in e-commerce: to predict whether two online user identities that access an e-commerce site from different devices belong to one real-world person. As input, we have a set of user activity logs from Taobao and some labeled user identity linkages. User activity logs can be modeled using a heterogeneous interaction graph (HIG), and subsequently the user alignment task can be formulated as a semi-supervised HIG embedding problem. HIG embedding is challenging for two reasons: its heterogeneous nature and the presence of edge features. To address the challenges, we propose a novel Heterogeneous Embedding Propagation (HEP) model. The core idea is to iteratively reconstruct a node's embedding from its heterogeneous neighbors in a weighted manner, and meanwhile propagate its embedding updates from reconstruction loss and/or classification loss to its neighbors. We conduct extensive experiments on large-scale datasets from Taobao, demonstrating that HEP significantly outperforms state-of-the-art baselines often by more than 10% in F-scores.

Network representation learning (RL) aims to transform the nodes in a network into low-dimensional vector spaces while preserving the inherent properties of the network. Though network RL has been intensively studied, most existing works focus on either network structure or node attribute information. In this paper, we propose a novel framework, named ANRL, to incorporate both the network structure and node attribute information in a principled way. Specifically, we propose a neighbor enhancement autoencoder to model the node attribute information, which reconstructs its target neighbors instead of itself. To capture the network structure, attribute-aware skip-gram model is designed based on the attribute encoder to formulate the correlations between each node and its direct or indirect neighbors. We conduct extensive experiments on six real-world networks, including two social networks, two citation networks and two user behavior networks. The results empirically show that ANRL can achieve relatively significant gains in node classification and link prediction tasks.

Machine learning (ML) systems have been increasingly applied in web security applications such as spammer detection, malware detection and fraud detection. These applications have an intrinsic adversarial nature where intelligent attackers can adaptively change their behaviors to avoid being detected by the deployed detectors. Existing efforts against adversaries are usually limited by the type of applied ML models or the specific applications such as image classification. Additionally, the working mechanisms of ML models usually cannot be well understood by users, which in turn impede them from understanding the vulnerabilities of models nor improving their robustness. To bridge the gap, in this paper, we propose to investigate whether model interpretation could potentially help adversarial detection. Specifically, we develop a novel adversary-resistant detection framework by utilizing the interpretation of ML models. The interpretation process explains the mechanism of how the target ML model makes prediction for a given instance, thus providing more insights for crafting adversarial samples. The robustness of detectors is then improved through adversarial training with the adversarial samples. A data-driven method is also developed to empirically estimate costs of adversaries in feature manipulation. Our approach is model-agnostic and can be applied to various types of classification models. Our experimental results on two real-world datasets demonstrate the effectiveness of interpretation-based attacks and how estimated feature manipulation cost would affect the behavior of adversaries.

Semantic proximity search on heterogeneous graph is an important task, and is useful for many applications. It aims to measure the proximity between two nodes on a heterogeneous graph w.r.t. some given semantic relation. Prior work often tries to measure the semantic proximity by paths connecting a query object and a target object. Despite the success of such path-based approaches, they often modeled the paths in a weakly coupled manner, which overlooked the rich interactions among paths. In this paper, we introduce a novel concept of interactive paths to model the inter-dependency among multiple paths between a query object and a target object. We then propose an Interactive Paths Embedding (IPE) model, which learns low-dimensional representations for the resulting interactive-paths structures for proximity estimation. We conduct experiments on seven relations with four different types of heterogeneous graphs, and show that our model outperforms the state-of-the-art baselines.

The e-commerce era is witnessing a rapid increase of mobile Internet users. Major e-commerce companies nowadays see billions of mobile accesses every day. Hidden in these records are valuable user behavioral characteristics such as their shopping preferences and browsing patterns. And, to extract these knowledge from the huge dataset, we need to first link records to the corresponding mobile devices. This Mobile Access Records Resolution (MARR) problem is confronted with two major challenges: (1) device identifiers and other attributes in access records might be missing or unreliable; (2) the dataset contains billions of access records from millions of devices. To the best of our knowledge, as a novel challenge industrial problem of mobile Internet, no existing method has been developed to resolve entities using mobile device identifiers in such a massive scale. To address these issues, we propose a SParse Identifier-linkage Graph (SPI-Graph) accompanied with the abundant mobile device profiling data to accurately match mobile access records to devices. Furthermore, two versions (unsupervised and semi-supervised) of Parallel Graph-based Record Resolution (PGRR) algorithm are developed to effectively exploit the advantages of the large-scale server clusters comprising of more than 1,000 computing nodes. We empirically show superior performances of PGRR algorithms in a very challenging and sparse real data set containing 5.28 million nodes and 31.06 million edges from 2.15 billion access records compared to other state-of-the-arts methodologies.

In the era of big data, it is often the rare categories that are of great interest in many high-impact applications, ranging from financial fraud detection in online transaction networks to emerging trend detection in social networks, from network intrusion detection in computer networks to fault detection in manufacturing. As a result, rare category characterization becomes a fundamental learning task, which aims to accurately characterize the rare categories given limited label information. The unique challenge of rare category characterization, i.e., the non-separability nature of the rare categories from the majority classes, together with the availability of the multi-modal representation of the examples, poses a new research question: how can we learn a salient rare category oriented embedding representation such that the rare examples are well separated from the majority class examples in the embedding space, which facilitates the follow-up rare category characterization?

Semantic user search is an important task on heterogeneous social networks. Its core problem is to measure the proximity between two user objects in the network w.r.t. certain semantic user relation. State-of-the-art solutions often take a path-based approach, which uses the sequences of objects connecting a query user and a target user to measure their proximity. Despite their success, we assert that path as a low-order structure is insufficient to capture the rich semantics between two users. Therefore, in this paper we introduce a new concept of subgraph-augmented path for semantic user search. Specifically, we consider sampling a set of object paths from a query user to a target user; then in each object path, we replace the linear object sequence between its every two neighboring users with their shared subgraph instances. Such subgraph-augmented paths are expected to leverage both path»s distance awareness and subgraph»s high-order structure. As it is non-trivial to model such subgraph-augmented paths, we develop a Subgraph-augmented Path Embedding (SPE) framework to accomplish the task. We evaluate our solution on six semantic user relations in three real-world public data sets, and show that it outperforms the baselines.

The social triad—a group of three people—is one of the simplest and most fundamental social groups. Extensive network and social theories have been developed to understand its structure, such as triadic closure and social balance. Over the course of a triadic closure—the transition from two ties to three among three users, the strength dynamics of its social ties, however, are much less well understood. Using two dynamic networks from social media and mobile communication, we examine how the formation of the third tie in a triad affects the strength of the existing two ties. Surprisingly, we find that in about 80% social triads, the strength of the first two ties is weakened although averagely the tie strength in the two networks maintains an increasing or stable trend. We discover that (1) the decrease in tie strength among three males is more sharply than that among females, and (2) the tie strength between celebrities is more likely to be weakened as the closure of a triad than those between ordinary people. Furthermore, we formalize a triadic tie strength dynamics prediction problem to infer whether social ties of a triad will become weakened after its closure. We propose a TRIST method—a kernel density estimation (KDE)-based graphical model—to solve the problem by incorporating user demographics, temporal effects, and structural information. Extensive experiments demonstrate that TRIST offers a greater than 82% potential predictability for inferring triadic tie strength dynamics in both networks. The leveraging of the KDE and structural correlations enables TRIST to outperform baselines by up to 30% in terms of F1-score.

Display Advertising has generated billions of revenue and originated hundreds of scientific papers and patents, yet the accuracy of prediction technologies leaves much to be desired. Conversion rates (CVR) predictions can often be formulated as a matrix or tensor completion problem where each dimension consists of thousands or even hundreds of thousands of levels. Observed entries are typically extremely sparse, comprising only 0.01% to 1% of the entire matrix or tensor with highly unevenly distributed conversion as well as impression sizes.

In this paper, we introduce a generic inference hybrid framework for Convolutional Recurrent Neural Network (conv-RNN) of semantic modeling of text, seamless integrating the merits on extracting different aspects of linguistic information from both convolutional and recurrent neural network structures and thus strengthening the semantic understanding power of the new framework. Besides, based on conv-RNN, we also propose a novel sentence classification model and an attention based answer selection model with strengthening power for the sentence matching and classification respectively. We validate the proposed models on a very wide variety of data sets, including two challenging tasks of answer selection (AS) and five benchmark datasets for sentence classification (SC). To the best of our knowledge, it is by far the most complete comparison results in both AS and SC. We empirically show superior performances of conv-RNN in these different challenging tasks and benchmark datasets and also summarize insights on the performances of other state-of-the-arts methodologies.

User behavior modeling is essential in computational advertisement, which builds users' profiles by tracking their online behaviors and then delivers the relevant ads according to each user's interests and needs. Accurate models will lead to higher targeting accuracy and thus improved advertising performance. Intuitively, similar users tend to have similar behaviors towards the displayed ads (e.g., impression, click, conversion). However, to the best of our knowledge, there is not much previous work that explicitly investigates such similarities of various types of user behaviors, and incorporates them into ad response targeting and prediction, largely due to the prohibitive scale of the problem.

We model the ad selection task as a multi-armed bandit problem. Standard assumptions in the multi-armed bandit (MAB) setting are that samples drawn from each arm areindependent and identically distributed, rewards (or conversionrates in our scenario) are stationary and rewards feedback areimmediate. Although the payoff function of an arm is allowed toevolve over time, the evolution is assumed to be slow. Display ads, on the other hand, are regularly created while others are removed from circulation. This can occur when budgets run out, campaign goal changes, holiday season ends and many other latent factors that go beyond the control of the ad selection system. Another big challenge is that the set of available ads is often extremely huge but standard multi-armed bandit strategies converge with linear time complexity that cannot accommodate the usually dynamic changes. Due to the above challenges and the restrictions of the original MAB, we propose a novel dynamic contextual MAB which tightly integrates components of dynamic conversion rates prediction, contextual learning and arm overlapping modeling in a principled framework. Besides we propose an accompaniedmeta analyses framework that allows us to conclude experiments in a more statistically robust manner. We demonstrate on a world leading demand side platform (DSP) that our framework can effectively discriminate premium arms and significantly outperform some standard variations of MAB to these settings.

Multiple types of heterogeneity including label heterogeneity and feature heterogeneity often co-exist in many real-world data mining applications, such as diabetes treatment classification, gene functionality prediction, and brain image analysis. To effectively leverage such heterogeneity, in this article, we propose a novel graph-based model for Learning with both Label and Feature heterogeneity, namely L 2 F . It models the label correlation by requiring that any two label-specific classifiers behave similarly on the same views if the associated labels are similar, and imposes the view consistency by requiring that view-based classifiers generate similar predictions on the same examples. The objective function for L 2 F is jointly convex. To solve the optimization problem, we propose an iterative algorithm, which is guaranteed to converge to the global optimum. One appealing feature of L 2 F is that it is capable of handling data with missing views and labels. Furthermore, we analyze its generalization performance based on Rademacher complexity, which sheds light on the benefits of jointly modeling the label and feature heterogeneity. Experimental results on various biomedical datasets show the effectiveness of the proposed approach.

This paper targets the problem of cargo pricing optimization in the air cargo business. Given the features associated with a pair of origination and destination, how can we simultaneously predict both the optimal price for the bid stage and the outcome of the transaction (win rate) in the decision stage? In addition, it is often the case that the matrix representing pairs of originations and destinations has a block structure, i.e., the originations and destinations can be co-clustered such that the predictive models are similar within the same co-cluster, and exhibit significant variation among different co-clusters. How can we uncover the co-clusters of originations and destinations while constructing the dual predictive models for the two stages?

Multiple types of heterogeneity, such as label heterogeneity and feature heterogeneity, often co-exist in many real-world data mining applications, such as news article categorization, gene functionality prediction. To effectively leverage such heterogeneity, in this paper, we propose a novel graph-based framework for Learning with both Label and Feature heterogeneities, namely L2F. It models the label correlation by requiring that any two label-specific classifiers behave similarly on the same views if the associated labels are similar, and imposes the view consistency by requiring that view-based classifiers generate similar predictions on the same examples. To solve the resulting optimization problem, we propose an iterative algorithm, which is guaranteed to converge to the global optimum. Furthermore, we analyze its generalization performance based on Rademacher complexity, which sheds light on the benefits of jointly modeling the label and feature heterogeneity. Experimental results on various data sets show the effectiveness of the proposed approach.

Traditional data mining techniques are designed to model a single type of heterogeneity, such as multi-task learning for modeling task heterogeneity, multi-view learning for modeling view heterogeneity, etc. Recently, a variety of real applications emerged, which exhibit dual heterogeneity, namely both task heterogeneity and view heterogeneity. Examples include insider threat detection across multiple organizations, web image classification in different domains, etc. Existing methods for addressing such problems typically assume that multiple tasks are equally related and multiple views are equally consistent, which limits their application in complex settings with varying task relatedness and view consistency. In this paper, we advance state-of-the-art techniques by adaptively modeling task relatedness and view consistency via a nonparametric Bayes model: we model task relatedness using normal penalty with sparse covariances, and view consistency using matrix Dirichlet process. Based on this model, we propose the NOBLE algorithm using an efficient Gibbs sampler. Experimental results on multiple real data sets demonstrate the effectiveness of the proposed algorithm.

A dependent hierarchical beta process (dHBP) is developed as a prior for data that may be represented in terms of a sparse set of latent features (dictionary elements), with covariate dependent feature usage. The dHBP is applicable to general covariates and data models, imposing that signals with similar covariates are likely to be manifested in terms of similar features. As an application, we consider the simultaneous sparse modeling of multiple images, with the covariate of a given image linked to its similarity to all other images (as applied in manifold learning). Efficient inference is performed using hybrid Gibbs, Metropolis-Hastings and slice sampling.