Spatio-Temporal Video Grounding (STVG) aims to retrieve the spatio-temporal tube of a target object or person in a video given a text query. Most existing approaches perform frame-wise spatial localization within a predicted temporal span, resulting in redundant computation, heavy supervision requirements, and limited generalization. Weakly-supervised variants mitigate annotation costs but remain constrained by the dataset-level train-and-fit paradigm with an inferior performance. To address these challenges, we propose the Agentic Spatio-Temporal Grounder (ASTG) framework for the task of STVG in an open-world and zero-shot setting. Specifically, two specialized agents SRA (Spatial Reasoning Agent) and TRA (Temporal Reasoning Agent) constructed leveraging modern Multi-modal Large Language Models (MLLMs) work collaboratively to retrieve the target tube in an autonomous and self-guided manner. Following a propose-and-evaluation paradigm, ASTG duly decouples spatio-temporal reasoning and automates the tube extraction, verification and temporal localization processes. With a dedicated visual memory and dialogue context, ASTG achieves architectural efficiency by minimizing the number of reasoning calls compared to exhaustive per-frame reasoning, eliminating the logical redundancy inherent in joint-reasoning systems. Experiments on popular benchmarks demonstrate the superiority of the proposed approach where it outperforms existing weakly-supervised and zero-shot approaches by a margin and is comparable to some of the fully-supervised methods.

Decentralized learning (DL) has proven to be effective for privacy-preserving next point-of-interest (POI) recommendation by sharing check-in information among users and collaboratively training on-device models. Existing works, however, simply assume that users tend to share check-ins with neighbors of short geographical distance or similar preferences yet ignore users’ actual willingness to share the information (WSI), causing potential privacy concerns. As such, we present a WSI-guided hierarchical DL framework for next POI recommendation (WHDL-Rec) to seek enhanced privacy protection with recommendation accuracy assured. In particular, WHDL-Rec first performs hierarchical data segregation to partition the private and public user data. It then accords to the server-client architecture, where the server exploits the public data to automatically learn users’ WSI w.r.t. check-ins and capture global user behavior patterns for recommendation accuracy maintenance; and the clients fuse the learned global patterns with the local private data for personalized on-device next POI recommendation, whereby WSI-guided collaborative learning is conducted with more secure check-in sharing. Extensive experiments on three real-world datasets demonstrate the efficacy of WHDL-Rec in delivering more accurate and privacy-preserved recommendations.

Vital signs, such as heart rate and blood pressure, are critical indicators of patient health and are widely used in clinical monitoring and decision-making. While deep learning models have shown promise in forecasting these signals, their deployment in healthcare remains limited in part because clinicians must be able to trust and interpret model outputs. Without reliable uncertainty quantification - particularly calibrated prediction intervals (PIs) - it is unclear whether a forecasted abnormality constitutes a meaningful warning or merely reflects model noise, hindering clinical decision-making. To address this, we present two methods for deriving PIs from the Reconstruction Uncertainty Estimate (RUE), an uncertainty measure well-suited to vital-sign forecasting due to its sensitivity to data shifts and support for label-free calibration. Our parametric approach assumes that prediction errors and uncertainty estimates follow a Gaussian copula distribution, enabling closed-form PI computation. Our non-parametric approach, based on k-nearest neighbours (KNN), empirically estimates the conditional error distribution using similar validation instances. We evaluate these methods on two large public datasets with minute- and hour-level sampling, representing high- and low-frequency health signals. Experiments demonstrate that the Gaussian copula method consistently outperforms conformal prediction baselines on low-frequency data, while the KNN approach performs best on high-frequency data. These results underscore the clinical promise of RUE-derived PIs for delivering interpretable, uncertainty-aware vital sign forecasts.

Contemporary Vision-Language Models (VLMs) such as CLIP offer an attractive zero-shot classification functionality facilitated by large-scale vision-language pre-training. However, they remain vulnerable to adversarial attacks, a critical security threat in realistic deployment. Adversarially robust fine-tuning provides generalizable robustness on new datasets while preserving natural performance by fine-tuning the pre-trained models. Fine-tuning robust CLIP typically relies on adversaries generated solely from the vision branch. However, this singular focus on the vision modality, coupled with static text prompts used as fixed category prototypes, limits the robustness achieved through dual-modality fine-tuning. We observe for CLIP fine-tuning that zero-shot adversarial robustness improves when we (i) stabilize the modality gap (a phenomenon where image and text features occupy different feature space regions) and (ii) lower/stabilize gradient norms. Both these steps enjoy further improvement of robustness if one fine-tunes with both visual and text adversaries. For both modalities, we leverage (i) the maximization of an effective rank of features and (ii) noise modulation of features. We show that maximizing the effective rank helps lower and stabilize the modality gap over adversaries with varying perturbation radii. The noise modulation of features, achieved by the so-called count sketching, lowers/stabilizes gradient norms. We outperform the state of the art on 15 datasets. We provide the first insights into the effects of modality gap & gradient norms in VLM fine-tuning.

Social recommendations (SRs) aim to enhance preference modeling by integrating social networks. However, their effectiveness is mainly constrained by two factors: the noisy social connections that may not reflect shared interests, and the limited number of social connections for most users, which hampers the system's ability to fully leverage social influence. Therefore, it is essential to perform social network refinement by removing noisy connections and adding meaningful ones for robust SRs. Inspired by the denoising capability of generative diffusion models, we propose a Model-Agnostic Social Network Refinement framework with Diffusion Models for Robust Social Recommendation (ARD-SR). Specifically, in the forward process, we corrupt the social network by progressively adding position-specific Gaussian noise calibrated to the user preference similarity, better simulating how the social network responds to noise perturbations. The reverse process learns to denoise, guided by each user's neighborhood preferences from the SR backbone, generating a tailored social network aligned with each user's preference for establishing connections. For effective learning, we design a curriculum-based training mechanism that progressively introduces challenging samples characterized by high sparsity or high noise levels. Finally, ARD-SR and the SR backbone are alternately trained, ensuring a continuous mutual enhancement between the social network refinement and the backbone's user representation learning. To further enhance the quality of the refined social network, (1) we introduce a preference-guided flip operation during inference to improve the input quality; and (2) we modify social connections based on the exponential weighted moving average of ARD-SR's predictions across epochs to reduce fluctuations. Experiments on three datasets show that ARD-SR significantly improves SR performance across multiple SR backbones. The code is released at https://github.com/sunyc123r/ARD-SR.

High-dimensional and incomplete (HDI) data are frequently encountered in diverse real-world applications involving complex interactions among numerous nodes. Approaches based on latent feature analysis (LFA) have proven effective in performing representation learning in HDI data. Nevertheless, they cannot handle the high-order connectivity among nodes in HDI data well, resulting in severe accuracy loss. To address the previously mentioned issue, we present a novel model in this paper, namely Graph Linear Convolution Pooling Network (GLCPN). The proposed GLCPN adopts the three-fold ideas. First, it leverages simplified graph convolutions to efficiently capture high-order connectivity among nodes for learning representations of matrix factorization. Second, a simple yet effective priori convolution operator is adopted by each graph neural layer to capture node-node collaboration for aggregation. Third, a locality-enhanced pooling scheme is designed to holistically utilize multi-layer representations of the neighborhood. Therefore, GLCPN can effectively acquire the hidden information in HDI data with high efficiency. In addition, we have conducted a theoretical analysis demonstrating that the proposed GLCPN is more expressive compared with existing graph neural networks for HDI data. Extensive experiments have been further conducted on ten well-established HDI datasets from various applications. The experimental results demonstrate that the proposed GLCPN significantly outperforms state-of-the-art models for learning representations in HDI data evaluated by accuracy and efficiency metrics.

The rapid growth of the internet has led to an alarming increase in the dissemination of fake news, which has had many negative effects on society. Various methods have been proposed for detecting fake news. However, these approaches suffer from several limitations. First, most existing works only consider news as separate entities and do not consider the correlations between fake news and real news. Moreover, these works are usually conducted in the Euclidean space, which is unable to capture complex relationships between news, in particular the hierarchical relationships. To tackle these issues, we introduce a novelMulti-modalHyperbolicRepresentation framework (MHR) for fake news detection. Specifically, we capture the correlations between news for graph construction to arrange and analyze different news. To fully utilize the multi-modal characteristics, we first extract the textual and visual information, and then design a Lorentzian multi-modal fusion module to fuse them as the node information in the graph. By utilizing the fully hyperbolic graph neural networks, we learn the graph’s representation in hyperbolic space, followed by a detector for detecting fake news. The experimental results on three real-world datasets demonstrate that our proposed MHR model achieves state-of-the-art performance, indicating the benefits of hyperbolic representation.

The next Point-of-interest recommendation has attracted extensive research interest recently, which predicts users' subsequent movements. The main challenge is how to effectively capture users' personalized sequential transitions in check-in trajectory, and various methods have been developed. However, most existing studies ignore the temporal information when conducting the next POI recommendation. To fill this gap, we investigate a time-specific next POI recommendation task, which additionally incorporates the target time information. We propose a brand new Time2Rotation technique to capture the temporal information. Different from conventional methods, we represent timeslots as rotation vectors and then perform the rotation operations. Based on the Time2Rotation technique, we propose a novel rotation-based temporal attention network, namely ROTAN, for the time-specific next POI recommendation task. The ROTAN begins by building a collaborative POI transition graph, capturing the asymmetric temporal influence in sequential transitions. After that, it incorporates temporal information into the modeling of individual check-in trajectories, extracting separate representations for user preference and POI influence to reflect their distinct temporal patterns. Lastly, the target time is integrated to generate recommendations. Extensive experiments are conducted on three real-world datasets, which demonstrates the advantages of the proposed Time2Rotation technique and ROTAN recommendation model.

Social Recommendation (SR) typically exploits neighborhood influence in the social network to enhance user preference modeling. However, users' intricate social behaviors may introduce noisy social connections for user modeling and harm the models' robustness. Existing solutions to alleviate social noise either filter out the noisy connections or generate new potential social connections. Due to the absence of labels, the former approaches may retain uncertain connections for user preference modeling while the latter methods may introduce additional social noise. Through data analysis, we discover that (1) social noise likely comes from the connected users with low preference similarity; and (2) Opinion Leaders (OLs) play a pivotal role in influence dissemination, surpassing high-similarity neighbors, regardless of their preference similarity with trusting peers. Guided by these observations, we propose a novel Self-Supervised Denoising approach through Independent Cascade Graph Augmentation, for more robust SR. Specifically, we employ the independent cascade diffusion model to generate an augmented graph view, which traverses the social graph and activates the edges in sequence to simulate the cascading influence spread. To steer the augmentation towards a denoised social graph, we (1) introduce a hierarchical contrastive loss to prioritize the activation of OLs first, followed by high-similarity neighbors, while weakening the low-similarity neighbors; and (2) integrate an information bottleneck based contrastive loss, aiming to minimize mutual information between original and augmented graphs yet preserve sufficient information for improved SR. Experiments conducted on two public datasets demonstrate that our model outperforms the state-of-the-art while also exhibiting higher robustness to different extents of social noise.

Long-term time series forecasting (LTSF) aims to predict future values for a long time based on historical data. The period term is an essential component of the time series, which is complex yet important for LTSF. Although existing studies have achieved promising results, they still have limitations in modeling dynamic complicated periods. Most studies only focus on static periods with fixed time steps, while very few studies attempt to capture dynamic periods in the time domain. In this paper, we dissect the original time series in time and frequency domains and empirically find that changes in periods are more easily captured and quantified in the frequency domain. Based on this observation, we propose to explore dynamic period features using rotation in the frequency domain. To this end, we develop the frequency-based rotation network (FRNet), a novel LTSF method to effectively capture the features of the dynamic complicated periods. FRNet decomposes the original time series into period and trend components. Based on the complex-valued linear networks, it leverages a period frequency rotation module to predict the period component and a patch frequency rotation module to predict the trend component, respectively. Extensive experiments on seven real-world datasets consistently demonstrate the superiority of FRNet over various state-of-the-art methods. The source code is available at https://github.com/SiriZhang45/FRNet.

Most existing denoising recommendation methods alleviate noisy implicit feedback (user behaviors) through mainly empirical studies. However, such studies may lack theoretical explainability and fail to model comprehensive noise patterns, which hinders the understanding and capturing of different noise patterns that affect users’ behaviors. Thus, we propose to capture comprehensive noise patterns through theoretical and empirical analysis for more effective denoising, where users’ behaviors are divided into willingness and action phases to disentangle independent noise patterns. Willingness refers to the user’s intent to interact with an item, which may not lead to actual interaction due to different factors such as misclicking. Action denotes the user’s actual interaction with an item. Our analysis unveils that (1) in the willingness phase, high uncertainty in the user’s willingness to interact with the item can lead to high expectation loss which aligns with the findings of existing denoising methods; and (2) in the action phase, higher user-specific inconsistency between willingness and action not only leads to more noise in the user’s overall behaviors but also makes it harder to distinguish between true and noisy behaviors. Inspired by these findings, we propose a Unified Denoising Training (UDT) method for recommendation. To alleviate uncertainty in the willingness phase, we lower the importance of the user-item interaction with high willingness uncertainty recognized by high loss. To ease the inconsistency in the action phase, we lower the importance for users with high user-specific inconsistency as it may lead to noisier behaviors. Then, we increase the importance gap between the clean and noisy behaviors for users with low user-specific inconsistency as their behaviors are more distinguishable. Extensive experiments on three real-world datasets show that our proposed UDT outperforms state-of-the-art denoising recommendation methods.

Most existing bundle generation approaches fall short in generating fixed-size bundles. Furthermore, they often neglect the underlying user intents reflected by the bundles in the generation process, resulting in less intelligible bundles. This paper addresses these limitations through the exploration of two interrelated tasks, i.e., personalized bundle generation and the underlying intent inference, based on different user sessions. Inspired by the reasoning capabilities of large language models (LLMs), we propose an adaptive in-context learning paradigm, which allows LLMs to draw tailored lessons from related sessions as demonstrations, enhancing the performance on target sessions. Specifically, we first employ retrieval augmented generation to identify nearest neighbor sessions, and then carefully design prompts to guide LLMs in executing both tasks on these neighbor sessions. To tackle reliability and hallucination challenges, we further introduce (1) a self-correction strategy promoting mutual improvements of the two tasks without supervision signals and (2) an auto-feedback mechanism for adaptive supervision based on the distinct mistakes made by LLMs on different neighbor sessions. Thereby, the target session can gain customized lessons for improved performance by observing the demonstrations of its neighbor sessions. Experiments on three real-world datasets demonstrate the effectiveness of our proposed method.

The goal of intent-aware session recommendation (ISR) approaches is to capture user intents within a session for accurate next-item prediction. However, the capability of these approaches is limited by assuming all sessions have a uniform and fixed number of intents. In reality, user sessions can vary, where the number of intentions may differ from one to another. Moreover, they can only learn user intents in the latent space, which further restricts the model's transparency. To ease these issues, we propose a simple yet effective paradigm for ISR motivated by the advanced reasoning capability of large language models (LLMs). Specifically, we first create an initial prompt to instruct LLMs to predict the next item by inferring varying user intents reflected in a session. Then, we propose an effective optimization mechanism to automatically optimize prompts with an iterative self-reflection. Finally, we leverage the robust generalizability of LLMs across diverse domains to efficiently select the optimal prompt for ISR. As such, the proposed paradigm effectively guides LLMs to identify varying user intents at a semantic level, thus delivering more accurate and comprehensible recommendations. Extensive experiments on three real-world datasets verify the superiority of our proposed method.

Graph clusters (or communities) represent important graph structural information. In this paper, we presentDifferentiableClustering for graphATtention (DCAT). To the best of our knowledge, DCAT is the first solution that incorporates graph clustering into graph attention networks (GAT) to learn cluster-aware attention scores for semi-supervised learning tasks. In DCAT, we propose a novel approach to formunderlineating graph clustering as an auxiliary differentiable objective based on modunderlinearity maximization, which can be optimized together with the learning objective of GAT for a semi-supervised task. Specifically, we propose a solution to relaxing modunderlinearity maximization from a discrete optimization problem to a differentiable objective with theoretical guarantee so that we can learn cluster-aware attention scores by jointly learning from graph clustering and a semi-supervised learning task. To address the computational challenge, we further propose to reformunderlineate the constraint introduced by the clustering objective into a new form. Our analysis shows that DCAT allocates higher attention scores to nodes within the same cluster, allowing them to have a higher influence in node representation learning, and thus DCAT will generate better node representations for downstream applications. The experimental resunderlinets on commonly used datasets show that DCAT outperforms popunderlinear and state-of-the-art graph neural networks.

Current study on next point-of-interest (POI) recommendation mainly explores user sequential transitions with the fine-grained individual-user POI check-in trajectories only, which suffers from the severe check-in data sparsity issue. In fact, coarse-grained signals (i.e., region- and global-level check-ins) in such sparse check-ins would also benefit to augment user preference learning. Specifically, our data analysis unveils that user movement exhibits noticeable patterns w.r.t. the regions of visited POIs. Meanwhile, the global all-user check-ins can help reflect sequential regularities shared by the crowd. We are, therefore, inspired to propose the MCMG: a Multi-Channel next POI recommendation framework with Multi-Granularity signals categorized from two orthogonal perspectives, i.e., fine-coarse grained check-ins at either POI/region level or local/global level. The MCMG is equipped with three modules, namely, global user behavior encoder, local multi-channel (i.e., region, category, and POI channels) encoder, and region-aware weighting strategy. Such design enables MCMG to be capable of capturing both fine- and coarse-grained sequential regularities as well as exploring the dynamic impact of multi-channel by differentiating the check-in patterns w.r.t. visited regions. Extensive experiments on four real-world datasets show that our MCMG significantly outperforms state-of-the-art next POI recommendation approaches.

Current session-based recommender systems (SBRSs) mainly focus on maximizing recommendation accuracy, while few studies have been devoted to improve diversity beyond accuracy. Meanwhile, it is unclear how the accuracy-oriented SBRSs perform in terms of diversity. In addition, the asserted “tradeoff” relationship between accuracy and diversity has been increasingly questioned in the literature. Toward the aforementioned issues, we conduct a holistic study to particularly examine the recommendation performance of representative SBRSs w.r.t. both accuracy and diversity, striving for better understanding of the diversity-related issues for SBRSs and providing guidance on designing diversified SBRSs. Particularly, for a fair and thorough comparison, we deliberately select state-of-the-art non-neural, deep neural, and diversified SBRSs by covering more scenarios with appropriate experimental setups, e.g., representative datasets, evaluation metrics, and hyper-parameter optimization technique. The source code can be obtained via github.com/qyin863/Understanding-Diversity-in-SBRSs . Our empirical results unveil that (1) non-diversified methods can also obtain satisfying performance on diversity, which can even surpass diversified ones, and (2) the relationship between accuracy and diversity is quite complex. Besides the “tradeoff” relationship, they can be positively correlated with each other, that is, having a same-trend (win–win or lose–lose) relationship, which varies across different methods and datasets. Additionally, we further identify three possible influential factors on diversity in SBRSs (i.e., granularity of item categorization, session diversity of datasets, and length of recommendation lists) and offer an intuitive guideline and a potential solution regarding learned item embeddings for more effective session-based recommendation.

Product bundling represents a prevalent marketing strategy in both offline stores and e-commerce systems. Despite its widespread use, previous studies on bundle recommendation face two significant limitations. Firstly, they rely on noisy datasets, where bundles are defined by heuristics, e.g., products co-purchased in the same session. Secondly, they target specific tasks by holding unrealistic assumptions, e.g., the availability of bundles for recommendation directly. This paper proposes to take a step back and considers the process of bundle recommendation from a holistic user experience perspective. We first construct high-quality bundle datasets with rich metadata, particularly bundle intents, through a carefully designed crowd-sourcing task. We then define a series of tasks that together, support all key steps in a typical bundle recommendation process, from bundle detection, completion and ranking, to explanation and auto-naming, whereby 19 research questions are raised correspondingly to guide the analysis. Finally, we conduct extensive experiments and analyses with representative recommendation models and large language models (LLMs), demonstrating the challenges and opportunities, especially with the emergence of LLMs. To summarize, our study contributes by introducing novel data sources, paving the way for new research avenues, and offering insights to guide product bundling in real e-commerce platforms.

Session-based recommendation (SR) has gained increasing attention in recent years. Quite a great amount of studies have been devoted to designing complex algorithms to improve recommendation performance, where deep learning methods account for the majority. However, most of these methods are black-box ones and ignore to provide moderate explanations to facilitate users’ understanding, which thus might lead to lowered user satisfaction and reduced system revenues. Therefore, in our study, we propose a generic Reinforced Explainable framework with Knowledge graph for Session-based recommendation (i.e., REKS), which strives to improve the existing black-box SR models (denoted as non-explainable ones) with Markov decision process. In particular, we construct a knowledge graph with session behaviors and treat SR models as part of the policy network of Markov decision process. Based on our particularly designed state vector, reward strategy, and loss function, the reinforcement learning (RL)-based framework not only achieves improved recommendation accuracy, but also provides appropriate explanations at the same time. Finally, we instantiate the REKS in five representative, state-of-the-art SR models (i.e., GRU4REC, NARM, SR-GNN, GCSAN, BERT4REC), whereby extensive experiments towards these methods on four datasets demonstrate the effectiveness of our framework on both recommendation and explanation tasks.

Item consumption and social connection, as common user behaviors in many web applications, have been extensively studied. However, most current works separately perform either item consumption or social link prediction tasks, possibly with the help of the other as an auxiliary signal. Moreover, they merely consider the behaviors in a holistic manner yet neglect the multi-faceted motives behind them. For example, the intention of watching a movie could be killing time or watching it with friends; Likewise, one might connect with others due to friendships or colleagues. To fill this gap, we propose to Disentangle the multi-faceted Motives in each network (i.e., the user-item interaction network and social network) defined respectively by the two types of behaviors, for mutually-enhanced Joint Prediction (DMJP). Specifically, we first learn the disentangled user representations driven by motives of multi-facets in both networks. Thereafter, the mutual influence of the two networks is subtly discriminated at the facet-to-facet level. The fine-grained mutual influence is then exploited asymmetrically to help refine user representations in both networks, with the goal of achieving a mutually-enhanced joint item and social link prediction. Empirical studies on three public datasets showcase the superiority of DMJP over state-of-the-arts (SOTAs) on both tasks.

Conversational recommender systems (CRSs) have become crucial emerging research topics in the field of RSs, thanks to their natural advantages of explicitly acquiring user preferences via interactive conversations and revealing the reasons behind recommendations. However, the majority of current CRSs are text-based, which is less user-friendly and may pose challenges for certain users, such as those with visual impairments or limited writing and reading abilities. Therefore,for the first time, this paper investigates the potential of voice-based CRS (VCRSs) to revolutionize the way users interact with RSs in a natural, intuitive, convenient, and accessible fashion. To support such studies, we create two VCRSs benchmark datasets in the e-commerce and movie domains, after realizing the lack of such datasets through an exhaustive literature review. Specifically, we first empirically verify the benefits and necessity of creating such datasets. Thereafter, we convert the user-item interactions to text-based conversations through the ChatGPT-driven prompts for generating diverse and natural templates, and then synthesize the corresponding audios via the text-to-speech model. Meanwhile, a number of strategies are delicately designed to ensure the naturalness and high quality of voice conversations. On this basis, we further explore the potential solutions and point out possible directions to build end-to-end VCRSs by seamlessly extracting and integrating voice-based inputs, thus delivering performance-enhanced, self-explainable, and user-friendly VCRSs. Our study aims to establish the foundation and motivate further pioneering research in the emerging field of VCRSs. This aligns with the principles of explainable AI and AI for social good, viz., utilizing technology's potential to create a fair, sustainable, and just world. Our codes and datasets are available on GitHub (https://github.com/hyllll/VCRS ).

Reading psychology believes text comprehension to involve a complex psychological construction process, with the reader mind being a dynamic associative system that stores an abundance of schemata. For Chinese text, in particular, the unique ideographic writing system allows its lansign to trigger semantic association and schema recalling without the need of phonetics. In contrast to previous research efforts on text classification problems, in this paper we present an interdisciplinary modeling approach that draws inspirations from the cognitive principles of ideography, schema theory and deep learning to study Chinese text classification. Specifically, we first propose a Radical-guided Associative Model (RAM) for preliminary cognitive imitation, which comprises two coupled spaces, namely the Literal Space and Associative Space. Then, taking consideration of the schemata acquired from the mind of a reader which plays a important role in influencing text-dependent information revision, we extend RAM with a systematic Schema-aware Radical-guided Associative Model (SRAM) that embeds label semantics as essential text-independent human knowledge for real-world abstraction. In SRAM, the Schema Space is introduced and a Schema Attention module is proposed with a novel loss paradigm that includes the linkage and interaction between text-dependent prior concepts and text-independent label schemata. Extensive experiments on three real-world datasets demonstrate the effectiveness and rationality of our proposed method.

Policy distillation (PD) has been widely studied in deep reinforcement learning (RL), while existing PD approaches assume that the demonstration data (i.e., state-action pairs in frames) in a decision making sequence is uniformly distributed. This may bring in unwanted bias since RL is a reward maximizing process instead of simple label matching. Given such an issue, we denote the frame importance as its contribution to the expected reward on a particular frame, and hypothesize that adapting such frame importance could benefit the performance of the distilled student policy. To verify our hypothesis, we analyze why and how frame importance matters in RL settings. Based on the analysis, we propose an importance prioritized PD framework that highlights the training on important frames, so as to learn efficiently. Particularly, the frame importance is measured by the reciprocal of weighted Shannon entropy from a teacher policy's action prescriptions. Experiments on Atari games and policy compression tasks show that capturing the frame importance significantly boosts the performance of the distilled policies.

Adversarial examples in automatic speech recognition (ASR) are naturally sounded by humans yet capable of fooling well trained ASR models to transcribe incorrectly. Existing audio adversarial examples are typically constructed by adding constrained perturbations on benign audio inputs. Such attacks are therefore generated with an audio dependent assumption. For the first time, we propose the Speech Synthesising based Attack (SSA), a novel threat model that constructs audio adversarial examples entirely from scratch, i.e., without depending on any existing audio to fool cutting-edge ASR models. To this end, we introduce a conditional variational auto-encoder (CVAE) as the speech synthesiser. Meanwhile, an adaptive sign gradient descent algorithm is proposed to solve the adversarial audio synthesis task. Experiments on three datasets (i.e., Audio Mnist, Common Voice, and Librispeech) show that our method could synthesise naturally sounded audio adversarial examples to mislead the start-of-the-art ASR models. Our web-page containing generated audio demos is at https://sites.google.com/view/ssa-asr/home.

Product bundling is a commonly-used marketing strategy in both offline retailers and online e-commerce systems. Current research on bundle recommendation is limited by: (1) noisy datasets, where bundles are defined by heuristics, e.g., products co-purchased in the same session; and (2) specific tasks, holding unrealistic assumptions, e.g., the availability of bundles for recommendation directly. In this paper, we propose to take a step back and consider the process of bundle recommendation from a holistic user experience perspective. We first construct high-quality bundle datasets with rich meta information, particularly bundle intents, through a carefully designed crowd-sourcing task. We then define a series of tasks that together, support all key steps in a typical bundle recommendation process, from bundle detection, completion, ranking, to explanation and auto-naming. Finally, we conduct extensive experiments and in-depth analysis that demonstrate the challenges of bundle recommendation, arising from the need for capturing complex relations among users, products, and bundles, as well as the research opportunities, especially in graph-based neural methods. To sum up, our study delivers new data sources, opens up new research directions, and provides useful guidance for product bundling in real e-commerce platforms. Our datasets are available at GitHub (\urlhttps://github.com/BundleRec/bundle_recommendation ).

Transfer covariance function, which can model domain similarity and adaptively control the knowledge transfer across domains, is widely used in transfer learning. In this paper, we concentrate on Gaussian process (GP) models using a transfer covariance function for regression problems in a black-box learning scenario. Precisely, we investigate a family of rather general transfer covariance functions,${T}_{*}$, that can model the heterogeneous sub-similarities of domains through multiple kernel learning. A necessary and sufficient condition to obtain validGPs using${T}_{*}$($GP_{T_{*}}$) for any data is given. This condition becomes specially handy for practical applications as (i) it enables semantic interpretations of the sub-similarities and (ii) it can readily be used for model learning. In particular, we propose a computationally inexpensive model learning rule that can explicitly capture different sub-similarities of domains. We propose two instantiations of$GP_{T_{*}}$, one with a set of predefined constant base kernels and one with a set of learnable parametric base kernels. Extensive experiments on 36 synthetic transfer tasks and 12 real-world transfer tasks demonstrate the effectiveness of$GP_{T_{*}}$on the sub-similarity capture and the transfer performance.

Sample-based evolutionary algorithms (EAs) are widely used for optimizing problems with multi (greater than one but less than four) or even many (greater than or equal to four) objectives of interest. In general, the difficulty of a problem exponentially increases with the number of objectives, serving as a clear example of the curse of dimensionality. The exploratory approach an EA takes in these cases has led to it being thought of as a big data generator, progressively sampling and evaluating solutions in high performing regions of a decision space to guide the search towards optimal solutions. Notably, in both multi- and many-objective EAs, the sampled data can be further utilized for building inverse generative models, mapping points in objective space back to solutions in the decision space. Such models offer immense flexibility to a decision maker in generating new target solutions on the fly, thereby facilitating real-time a posteriori preference incorporation into the search. In this paper, we show that the data distribution resulting from a many-objective formulation is in fact more conducive to building accurate inverse models than its multiobjective counterpart. Given the potential utility of these models, we in turn shed light on a rare blessing of dimensionality that is yet to be explored in the context of optimization. We first present simple theoretical arguments supporting our claim. Thereafter, experimental studies of Gaussian process-based inverse modeling for a synthetic and a real-world example are carried out to further confirm the theory.

Transfer optimization is an emerging concept that promises to enhance productivity of planning and decision-making processes by allowing for the adaptive reuse of knowledge (data) drawn from various “source” problems in a related ongoing “target” task of interest. Despite the recent advances in transfer optimization, however, a continuing challenge is the scalability of associated algorithms given big data of source problem instances. This paper tackles the scaling problem of an online adaptive knowledge transfer framework under big source data. We propose an efficient source selection algorithm based on the theory of multi-armed bandits such that the most related source task to the target is chosen for knowledge transfer, as opposed to extracting knowledge from all sources simultaneously. For this purpose, we introduce a novel and principled reward measure to reflect the source-target similarities. The efficacy of our proposed approach is assessed on the well-known knapsack problem that has practical implications in optimization of supply chain and manufacturing processes. Extensive experiments are conducted under big data of source problem instances. The numerical results clearly reveal that the incorporation of the proposed source selection mechanism in the existing adaptive knowledge transfer framework makes it successfully feasible for fast/real-time decision-making in the big data source setting.

Autonomous construction of deep neural network (DNNs) is desired for data streams because it potentially offers two advantages: proper model's capacity and quick reaction to drift and shift. While self-organizing mechanism of DNNs remains an open issue, this task is even more challenging to be developed for standard multi-layer DNNs than that using the different-depth structures, because addition of a new layer results in information loss of previously trained knowledge. A Neural Network with Dynamically Evolved Capacity (NADINE) is proposed in this paper. NADINE features a fully open structure where its network structure, depth and width, can be automatically evolved from scratch in the online manner and without the use of problem-specific thresholds. NADINE is structured under a standard MLP architecture and the catastrophic forgetting issue during the hidden layer addition phase is resolved using the proposal of soft-forgetting and adaptive memory methods. The advantage of NADINE, namely elastic structure and online learning trait, is numerically validated using nine data stream classification and regression problems where it demonstrates performance's improvement over prominent algorithms in all problems. In addition, it is capable of dealing with data stream regression and classification problems equally well.

Transfer covariance functions, which can model domain similarities and adaptively control the knowledge transfer across domains, are widely used in Gaussian process (GP) based transfer learning. We focus on regression problems in a black-box learning scenario, and study a family of rather general transfer covariance functions, T_*, that can model the similarity heterogeneity of domains through multiple kernel learning. A necessary and sufficient condition that (i) validates GPs using T_* for any data and (ii) provides semantic interpretations is given. Moreover, building on this condition, we propose a computationally inexpensive model learning rule that can explicitly capture different sub-similarities of domains. Extensive experiments on one synthetic dataset and four real-world datasets demonstrate the effectiveness of the learned GP on the sub-similarity capture and the transfer performance.

Evolutionary data, such as topic changing blogs and evolving trading behaviors in capital market, is widely seen in business and social applications. The time factor and intrinsic change embedded in evolutionary data greatly challenge evolutionary clustering. To incorporate the time factor, existing methods mainly regard the evolutionary clustering problem as a linear combination of snapshot cost and temporal cost, and reflect the time factor through the temporal cost. It still faces accuracy and scalability challenge though promising results gotten. This paper proposes a novel evolutionary clustering approach, evolutionary maximum margin clustering (e-MMC), to cluster large-scale evolutionary data from the maximum margin perspective. e-MMC incorporates two frameworks: Data Integration from the data changing perspective and Model Integration corresponding to model adjustment to tackle the time factor and change, with an adaptive label allocation mechanism. Three e-MMC clustering algorithms are proposed based on the two frameworks. Extensive experiments are performed on synthetic data, UCI data and real-world blog data, which confirm that e-MMC outperforms the state-of-the-art clustering algorithms in terms of accuracy, computational cost and scalability. It shows that e-MMC is particularly suitable for clustering large-scale evolving data.

In the absence of the labeled samples in a domain referred to as target domain, Domain Adaptation (DA) techniques come in handy. Generally, DA techniques assume there are available source domains that share similar predictive function with the target domain. Two core challenges of DA typically arise, variance that exists between source and target domains, and the inherent source hypothesis bias. In this paper, we first propose a Stability Transfer criterion for selecting relevant source domains with low variance. With this criterion, we introduce a TARget learning Assisted by Source Classifier Adaptation (TARASCA) method to address the two core challenges that have impeded the performances of DA techniques. To verify the robustness of TARASCA, extensive experimental studies are carried out with comparison to several state-of-the-art DA methods on the real-world Sentiment and Newsgroups datasets, where various settings for the class ratios of the source and target domains are considered.

Domain Adaptation (DA) methods are usually carried out by means of simply reducing the marginal distribution differences between the source and target domains, and subsequently using the resultant trained classifier, namely source classifier, for use in the target domain. However, in many cases, the true predictive distributions of the source and target domains can be vastly different especially when their class distributions are skewed, causing the issues of sample selection bias in DA. Hence, DA methods which leverage the source labeled data may suffer from poor generalization in the target domain, resulting in negative transfer. In addition, we observed that many DA methods use either a source classifier or a linear combination of source classifiers with a fixed weighting for predicting the target unlabeled data. Essentially, the labels of the target unlabeled data are spanned by the prediction of these source classifiers. Motivated by these observations, in this paper, we propose to construct many source classifiers of diverse biases and learn the weight for each source classifier by directly minimizing the structural risk defined on the target unlabeled data so as to heal the possible sample selection bias. Since the weights are learned by maximizing the margin of separation between opposite classes on the target unlabeled data, the proposed method is established here as Maximal Margin Target Label Learning (MMTLL), which is in a form of Multiple Kernel Learning problem with many label kernels. Extensive experimental studies of MMTLL against several state-of-the-art methods on the Sentiment and Newsgroups datasets with various imbalanced class settings showed that MMTLL exhibited robust accuracies on all the settings considered and was resilient to negative transfer, in contrast to other counterpart methods which suffered significantly in prediction accuracy.