Wynne Hsu

Jay Zhangjie Wu, Yixiao Ge, Xintao Wang et al. · tencent-ai

To replicate the success of text-to-image (T2I) generation, recent works employ large-scale video datasets to train a text-to-video (T2V) generator. Despite their promising results, such paradigm is computationally expensive. In this work, we propose a new T2V generation setting$\unicode{x2014}$One-Shot Video Tuning, where only one text-video pair is presented. Our model is built on state-of-the-art T2I diffusion models pre-trained on massive image data. We make two key observations: 1) T2I models can generate still images that represent verb terms; 2) extending T2I models to generate multiple images concurrently exhibits surprisingly good content consistency. To further learn continuous motion, we introduce Tune-A-Video, which involves a tailored spatio-temporal attention mechanism and an efficient one-shot tuning strategy. At inference, we employ DDIM inversion to provide structure guidance for sampling. Extensive qualitative and numerical experiments demonstrate the remarkable ability of our method across various applications.

Jay Zhangjie Wu, David Junhao Zhang, Wynne Hsu et al.

Humans can watch a continuous video stream and effortlessly perform continual acquisition and transfer of new knowledge with minimal supervision yet retaining previously learnt experiences. In contrast, existing continual learning (CL) methods require fully annotated labels to effectively learn from individual frames in a video stream. Here, we examine a more realistic and challenging problem$\unicode{x2014}$Label-Efficient Online Continual Object Detection (LEOCOD) in streaming video. We propose a plug-and-play module, Efficient-CLS, that can be easily inserted into and improve existing continual learners for object detection in video streams with reduced data annotation costs and model retraining time. We show that our method has achieved significant improvement with minimal forgetting across all supervision levels on two challenging CL benchmarks for streaming real-world videos. Remarkably, with only 25% annotated video frames, our method still outperforms the base CL learners, which are trained with 100% annotations on all video frames. The data and source code will be publicly available at https://github.com/showlab/Efficient-CLS.

Meng Luo, Hao Fei, Bobo Li et al.

While existing Aspect-based Sentiment Analysis (ABSA) has received extensive effort and advancement, there are still gaps in defining a more holistic research target seamlessly integrating multimodality, conversation context, fine-granularity, and also covering the changing sentiment dynamics as well as cognitive causal rationales. This paper bridges the gaps by introducing a multimodal conversational ABSA, where two novel subtasks are proposed: 1) Panoptic Sentiment Sextuple Extraction, panoramically recognizing holder, target, aspect, opinion, sentiment, rationale from multi-turn multi-party multimodal dialogue. 2) Sentiment Flipping Analysis, detecting the dynamic sentiment transformation throughout the conversation with the causal reasons. To benchmark the tasks, we construct PanoSent, a dataset annotated both manually and automatically, featuring high quality, large scale, multimodality, multilingualism, multi-scenarios, and covering both implicit and explicit sentiment elements. To effectively address the tasks, we devise a novel Chain-of-Sentiment reasoning framework, together with a novel multimodal large language model (namely Sentica) and a paraphrase-based verification mechanism. Extensive evaluations demonstrate the superiority of our methods over strong baselines, validating the efficacy of all our proposed methods. The work is expected to open up a new era for the ABSA community, and thus all our codes and data are open at https://PanoSent.github.io/

Evelyn Chee, Mong Li Lee, Wynne Hsu

Class-incremental continual learning is a core step towards developing artificial intelligence systems that can continuously adapt to changes in the environment by learning new concepts without forgetting those previously learned. This is especially needed in the medical domain where continually learning from new incoming data is required to classify an expanded set of diseases. In this work, we focus on how old knowledge can be leveraged to learn new classes without catastrophic forgetting. We propose a framework that comprises of two main components: (1) a dynamic architecture with expanding representations to preserve previously learned features and accommodate new features; and (2) a training procedure alternating between two objectives to balance the learning of new features while maintaining the model's performance on old classes. Experiment results on multiple medical datasets show that our solution is able to achieve superior performance over state-of-the-art baselines in terms of class accuracy and forgetting.

Haodi Jiang, Qin Li, Yan Xu et al.

Obtaining high-quality magnetic and velocity fields through Stokes inversion is crucial in solar physics. In this paper, we present a new deep learning method, named Stacked Deep Neural Networks (SDNN), for inferring line-of-sight (LOS) velocities and Doppler widths from Stokes profiles collected by the Near InfraRed Imaging Spectropolarimeter (NIRIS) on the 1.6 m Goode Solar Telescope (GST) at the Big Bear Solar Observatory (BBSO). The training data of SDNN is prepared by a Milne-Eddington (ME) inversion code used by BBSO. We quantitatively assess SDNN, comparing its inversion results with those obtained by the ME inversion code and related machine learning (ML) algorithms such as multiple support vector regression, multilayer perceptrons and a pixel-level convolutional neural network. Major findings from our experimental study are summarized as follows. First, the SDNN-inferred LOS velocities are highly correlated to the ME-calculated ones with the Pearson product-moment correlation coefficient being close to 0.9 on average. Second, SDNN is faster, while producing smoother and cleaner LOS velocity and Doppler width maps, than the ME inversion code. Third, the maps produced by SDNN are closer to ME's maps than those from the related ML algorithms, demonstrating the better learning capability of SDNN than the ML algorithms. Finally, comparison between the inversion results of ME and SDNN based on GST/NIRIS and those from the Helioseismic and Magnetic Imager on board the Solar Dynamics Observatory in flare-prolific active region NOAA 12673 is presented. We also discuss extensions of SDNN for inferring vector magnetic fields with empirical evaluation.

Haodi Jiang, Qin Li, Zhihang Hu et al.

Solar activity is usually caused by the evolution of solar magnetic fields. Magnetic field parameters derived from photospheric vector magnetograms of solar active regions have been used to analyze and forecast eruptive events such as solar flares and coronal mass ejections. Unfortunately, the most recent solar cycle 24 was relatively weak with few large flares, though it is the only solar cycle in which consistent time-sequence vector magnetograms have been available through the Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory (SDO) since its launch in 2010. In this paper, we look into another major instrument, namely the Michelson Doppler Imager (MDI) on board the Solar and Heliospheric Observatory (SOHO) from 1996 to 2010. The data archive of SOHO/MDI covers more active solar cycle 23 with many large flares. However, SOHO/MDI data only has line-of-sight (LOS) magnetograms. We propose a new deep learning method, named MagNet, to learn from combined LOS magnetograms, Bx and By taken by SDO/HMI along with H-alpha observations collected by the Big Bear Solar Observatory (BBSO), and to generate vector components Bx' and By', which would form vector magnetograms with observed LOS data. In this way, we can expand the availability of vector magnetograms to the period from 1996 to present. Experimental results demonstrate the good performance of the proposed method. To our knowledge, this is the first time that deep learning has been used to generate photospheric vector magnetograms of solar active regions for SOHO/MDI using SDO/HMI and H-alpha data.

Yixuan Tang, Zhenghong Lin, Yandong Sun et al.

While dense biomedical embeddings achieve strong performance, their black-box nature limits their utility in clinical decision-making. Recent question-based interpretable embeddings represent text as binary answers to natural-language questions, but these approaches often rely on heuristic or surface-level contrastive signals and overlook specialized domain knowledge. We propose QIME, an ontology-grounded framework for constructing interpretable medical text embeddings in which each dimension corresponds to a clinically meaningful yes/no question. By conditioning on cluster-specific medical concept signatures, QIME generates semantically atomic questions that capture fine-grained distinctions in biomedical text. Furthermore, QIME supports a training-free embedding construction strategy that eliminates per-question classifier training while further improving performance. Experiments across biomedical semantic similarity, clustering, and retrieval benchmarks show that QIME consistently outperforms prior interpretable embedding methods and substantially narrows the gap to strong black-box biomedical encoders, while providing concise and clinically informative explanations.

Svetlana Churina, Kokil Jaidka, Anab Maulana Barik et al.

The web's information ecosystem demands fact-checking systems that are both scalable and epistemically trustworthy. Automated approaches offer efficiency but often lack transparency, while human verification remains slow and inconsistent. We introduce Althea, a retrieval-augmented system that integrates question generation, evidence retrieval, and structured reasoning to support user-driven evaluation of online claims. On the AVeriTeC benchmark, Althea achieves a Macro-F1 of 0.44, outperforming standard verification pipelines and improving discrimination between supported and refuted claims. We further evaluate Althea through a controlled user study and a longitudinal survey experiment (N = 642), comparing three interaction modes that vary in the degree of scaffolding: an Exploratory mode with guided reasoning, a Summary mode providing synthesized verdicts, and a Self-search mode that offers procedural guidance without algorithmic intervention. Results show that guided interaction produces the strongest immediate gains in accuracy and confidence, while self-directed search yields the most persistent improvements over time. This pattern suggests that performance gains are not driven solely by effort or exposure, but by how cognitive work is structured and internalized.

Evelyn Chee, Wynne Hsu, Mong Li Lee

Continual learning is essential for adapting models to new tasks while retaining previously acquired knowledge. While existing approaches predominantly focus on uni-modal data, multi-modal learning offers substantial benefits by utilizing diverse sensory inputs, akin to human perception. However, multi-modal continual learning presents additional challenges, as the model must effectively integrate new information from various modalities while preventing catastrophic forgetting. In this work, we propose a pre-trained model-based framework for multi-modal continual learning. Our framework includes a novel cross-modality adapter with a mixture-of-experts structure to facilitate effective integration of multi-modal information across tasks. We also introduce a representation alignment loss that fosters learning of robust multi-modal representations, and regularize relationships between learned representations to preserve knowledge from previous tasks. Experiments on several multi-modal datasets demonstrate that our approach consistently outperforms baselines in both class-incremental and domain-incremental learning, achieving higher accuracy and reduced forgetting.

Yanguang Zhao, Jie Yang, Shengqiong Wu et al.

Spatial reasoning, the ability to understand spatial relations, causality, and dynamic evolution, is central to human intelligence and essential for real-world applications such as autonomous driving and robotics. Existing studies, however, primarily assess models on visible spatio-temporal understanding, overlooking their ability to infer unseen past or future spatial states. In this work, we introduce Spatial Causal Prediction (SCP), a new task paradigm that challenges models to reason beyond observation and predict spatial causal outcomes. We further construct SCP-Bench, a benchmark comprising 2,500 QA pairs across 1,181 videos spanning diverse viewpoints, scenes, and causal directions, to support systematic evaluation. Through comprehensive experiments on {23} state-of-the-art models, we reveal substantial gaps between human and model performance, limited temporal extrapolation, and weak causal grounding. We further analyze key factors influencing performance and propose perception-enhancement and reasoning-guided strategies toward advancing spatial causal intelligence. The project page is https://guangstrip.github.io/SCP-Bench.

Jundong Xu, Hao Fei, Meng Luo et al.

In the context of large language models (LLMs), current advanced reasoning methods have made impressive strides in various reasoning tasks. However, when it comes to logical reasoning tasks, major challenges remain in both efficacy and efficiency. This is rooted in the fact that these systems fail to fully leverage the inherent structure of logical tasks throughout the reasoning processes such as decomposition, search, and resolution. To address this, we propose a logic-complete reasoning framework, Aristotle, with three key components: Logical Decomposer, Logical Search Router, and Logical Resolver. In our framework, symbolic expressions and logical rules are comprehensively integrated into the entire reasoning process, significantly alleviating the bottlenecks of logical reasoning, i.e., reducing sub-task complexity, minimizing search errors, and resolving logical contradictions. The experimental results on several datasets demonstrate that Aristotle consistently outperforms state-of-the-art reasoning frameworks in both accuracy and efficiency, particularly excelling in complex logical reasoning scenarios. We will open-source all our code at https://llm-symbol.github.io/Aristotle/.

Tianjie Ju, Zhenyu Shao, Bowen Wang et al.

Large Language Models (LLMs) have demonstrated promising capabilities to generate responses that simulate consistent personality traits. Despite the major attempts to analyze personality expression through output-based evaluations, little is known about how such traits are internally encoded within LLM parameters. In this paper, we introduce a layer-wise probing framework to systematically investigate the layer-wise capability of LLMs in simulating personality for responding. We conduct probing experiments on 11 open-source LLMs over the PersonalityEdit benchmark and find that LLMs predominantly simulate personality for responding in their middle and upper layers, with instruction-tuned models demonstrating a slightly clearer separation of personality traits. Furthermore, by interpreting the trained probing hyperplane as a layer-wise boundary for each personality category, we propose a layer-wise perturbation method to edit the personality expressed by LLMs during inference. Our results show that even when the prompt explicitly specifies a particular personality, our method can still successfully alter the response personality of LLMs. Interestingly, the difficulty of converting between certain personality traits varies substantially, which aligns with the representational distances in our probing experiments. Finally, we conduct a comprehensive MMLU benchmark evaluation and time overhead analysis, demonstrating that our proposed personality editing method incurs only minimal degradation in general capabilities while maintaining low training costs and acceptable inference latency. Our code is publicly available at https://github.com/universe-sky/probing-then-editing-personality.

Meng Luo, Shengqiong Wu, Liqiang Jing et al.

Recent advancements in large video models (LVMs) have significantly enhance video understanding. However, these models continue to suffer from hallucinations, producing content that conflicts with input videos. To address this issue, we propose Dr.V, a hierarchical framework covering perceptive, temporal, and cognitive levels to diagnose video hallucination by fine-grained spatial-temporal grounding. Dr.V comprises of two key components: a benchmark dataset Dr.V-Bench and a satellite video agent Dr.V-Agent. Dr.V-Bench includes 10k instances drawn from 4,974 videos spanning diverse tasks, each enriched with detailed spatial-temporal annotation. Dr.V-Agent detects hallucinations in LVMs by systematically applying fine-grained spatial-temporal grounding at the perceptive and temporal levels, followed by cognitive level reasoning. This step-by-step pipeline mirrors human-like video comprehension and effectively identifies hallucinations. Extensive experiments demonstrate that Dr.V-Agent is effective in diagnosing hallucination while enhancing interpretability and reliability, offering a practical blueprint for robust video understanding in real-world scenarios. All our data and code are available at https://github.com/Eurekaleo/Dr.V.

Tianjie Ju, Yi Hua, Hao Fei et al.

Multi-Modal Large Language Models (MLLMs) have exhibited remarkable performance on various vision-language tasks such as Visual Question Answering (VQA). Despite accumulating evidence of privacy concerns associated with task-relevant content, it remains unclear whether MLLMs inadvertently memorize private content that is entirely irrelevant to the training tasks. In this paper, we investigate how randomly generated task-irrelevant private content can become spuriously correlated with downstream objectives due to partial mini-batch training dynamics, thus causing inadvertent memorization. Concretely, we randomly generate task-irrelevant watermarks into VQA fine-tuning images at varying probabilities and propose a novel probing framework to determine whether MLLMs have inadvertently encoded such content. Our experiments reveal that MLLMs exhibit notably different training behaviors in partial mini-batch settings with task-irrelevant watermarks embedded. Furthermore, through layer-wise probing, we demonstrate that MLLMs trigger distinct representational patterns when encountering previously seen task-irrelevant knowledge, even if this knowledge does not influence their output during prompting. Our code is available at https://github.com/illusionhi/ProbingPrivacy.

Tianjie Ju, Bowen Wang, Hao Fei et al.

Recent advances in Large Language Models (LLMs) have upgraded them from sophisticated text generators to autonomous agents capable of cooperation and tool use in multi-agent systems (MAS). However, it remains unclear how disagreements shape collective decision-making. In this paper, we revisit the role of disagreement and argue that general, partially overlapping disagreements prevent premature consensus and expand the explored solution space, while disagreements on task-critical steps can derail collaboration depending on the topology of solution paths. We investigate two collaborative settings with distinct path structures: collaborative reasoning (CounterFact, MQuAKE-cf), which typically follows a single evidential chain, whereas collaborative programming (HumanEval, GAIA) often adopts multiple valid implementations. Disagreements are instantiated as general heterogeneity among agents and as task-critical counterfactual knowledge edits injected into context or parameters. Experiments reveal that general disagreements consistently improve success by encouraging complementary exploration. By contrast, task-critical disagreements substantially reduce success on single-path reasoning, yet have a limited impact on programming, where agents can choose alternative solutions. Trace analyses show that MAS frequently bypasses the edited facts in programming but rarely does so in reasoning, revealing an emergent self-repair capability that depends on solution-path rather than scale alone. Our code is available at https://github.com/wbw625/MultiAgentRobustness.

Hao Fei, Shengqiong Wu, Wei Ji et al.

Existing research of video understanding still struggles to achieve in-depth comprehension and reasoning in complex videos, primarily due to the under-exploration of two key bottlenecks: fine-grained spatial-temporal perceptive understanding and cognitive-level video scene comprehension. This paper bridges the gap by presenting a novel solution. We first introduce a novel video Multimodal Large Language Model (MLLM), MotionEpic, which achieves fine-grained pixel-level spatial-temporal video grounding by integrating video spatial-temporal scene graph (STSG) representation. Building upon MotionEpic, we then develop a Video-of-Thought (VoT) reasoning framework. VoT inherits the Chain-of-Thought (CoT) core, breaking down a complex task into simpler and manageable sub-problems, and addressing them step-by-step from a low-level pixel perception to high-level cognitive interpretation. Extensive experiments across various complex video QA benchmarks demonstrate that our overall framework strikingly boosts existing state-of-the-art. To our knowledge, this is the first attempt at successfully implementing the CoT technique for achieving human-level video reasoning, where we show great potential in extending it to a wider range of video understanding scenarios. Project is open at https://haofei.vip/VoT

Peng Qi, Zehong Yan, Wynne Hsu et al.

Misinformation is a prevalent societal issue due to its potential high risks. Out-of-context (OOC) misinformation, where authentic images are repurposed with false text, is one of the easiest and most effective ways to mislead audiences. Current methods focus on assessing image-text consistency but lack convincing explanations for their judgments, which is essential for debunking misinformation. While Multimodal Large Language Models (MLLMs) have rich knowledge and innate capability for visual reasoning and explanation generation, they still lack sophistication in understanding and discovering the subtle crossmodal differences. In this paper, we introduce SNIFFER, a novel multimodal large language model specifically engineered for OOC misinformation detection and explanation. SNIFFER employs two-stage instruction tuning on InstructBLIP. The first stage refines the model's concept alignment of generic objects with news-domain entities and the second stage leverages language-only GPT-4 generated OOC-specific instruction data to fine-tune the model's discriminatory powers. Enhanced by external tools and retrieval, SNIFFER not only detects inconsistencies between text and image but also utilizes external knowledge for contextual verification. Our experiments show that SNIFFER surpasses the original MLLM by over 40% and outperforms state-of-the-art methods in detection accuracy. SNIFFER also provides accurate and persuasive explanations as validated by quantitative and human evaluations.

Jay Zhangjie Wu, Guian Fang, Haoning Wu et al. · tencent-ai

Generative models have demonstrated remarkable capability in synthesizing high-quality text, images, and videos. For video generation, contemporary text-to-video models exhibit impressive capabilities, crafting visually stunning videos. Nonetheless, evaluating such videos poses significant challenges. Current research predominantly employs automated metrics such as FVD, IS, and CLIP Score. However, these metrics provide an incomplete analysis, particularly in the temporal assessment of video content, thus rendering them unreliable indicators of true video quality. Furthermore, while user studies have the potential to reflect human perception accurately, they are hampered by their time-intensive and laborious nature, with outcomes that are often tainted by subjective bias. In this paper, we investigate the limitations inherent in existing metrics and introduce a novel evaluation pipeline, the Text-to-Video Score (T2VScore). This metric integrates two pivotal criteria: (1) Text-Video Alignment, which scrutinizes the fidelity of the video in representing the given text description, and (2) Video Quality, which evaluates the video's overall production caliber with a mixture of experts. Moreover, to evaluate the proposed metrics and facilitate future improvements on them, we present the TVGE dataset, collecting human judgements of 2,543 text-to-video generated videos on the two criteria. Experiments on the TVGE dataset demonstrate the superiority of the proposed T2VScore on offering a better metric for text-to-video generation.

Yingtian Zou, Kenji Kawaguchi, Yingnan Liu et al.

Generalizing to out-of-distribution (OOD) data or unseen domain, termed OOD generalization, still lacks appropriate theoretical guarantees. Canonical OOD bounds focus on different distance measurements between source and target domains but fail to consider the optimization property of the learned model. As empirically shown in recent work, the sharpness of learned minima influences OOD generalization. To bridge this gap between optimization and OOD generalization, we study the effect of sharpness on how a model tolerates data change in domain shift which is usually captured by "robustness" in generalization. In this paper, we give a rigorous connection between sharpness and robustness, which gives better OOD guarantees for robust algorithms. It also provides a theoretical backing for "flat minima leads to better OOD generalization". Overall, we propose a sharpness-based OOD generalization bound by taking robustness into consideration, resulting in a tighter bound than non-robust guarantees. Our findings are supported by the experiments on a ridge regression model, as well as the experiments on deep learning classification tasks.

Yingnan Liu, Yingtian Zou, Rui Qiao et al.

Domain generalization aims to develop models that are robust to distribution shifts. Existing methods focus on learning invariance across domains to enhance model robustness, and data augmentation has been widely used to learn invariant predictors, with most methods performing augmentation in the input space. However, augmentation in the input space has limited diversity whereas in the feature space is more versatile and has shown promising results. Nonetheless, feature semantics is seldom considered and existing feature augmentation methods suffer from a limited variety of augmented features. We decompose features into class-generic, class-specific, domain-generic, and domain-specific components. We propose a cross-domain feature augmentation method named XDomainMix that enables us to increase sample diversity while emphasizing the learning of invariant representations to achieve domain generalization. Experiments on widely used benchmark datasets demonstrate that our proposed method is able to achieve state-of-the-art performance. Quantitative analysis indicates that our feature augmentation approach facilitates the learning of effective models that are invariant across different domains.

Shenghan Wu, Yimo Zhu, Wynne Hsu et al.

The rapid advancement of Large Language Models (LLMs) has revolutionized the generation of emotional support conversations (ESC), offering scalable solutions with reduced costs and enhanced data privacy. This paper explores the role of personas in the creation of ESC by LLMs. Our research utilizes established psychological frameworks to measure and infuse persona traits into LLMs, which then generate dialogues in the emotional support scenario. We conduct extensive evaluations to understand the stability of persona traits in dialogues, examining shifts in traits post-generation and their impact on dialogue quality and strategy distribution. Experimental results reveal several notable findings: 1) LLMs can infer core persona traits, 2) subtle shifts in emotionality and extraversion occur, influencing the dialogue dynamics, and 3) the application of persona traits modifies the distribution of emotional support strategies, enhancing the relevance and empathetic quality of the responses. These findings highlight the potential of persona-driven LLMs in crafting more personalized, empathetic, and effective emotional support dialogues, which has significant implications for the future design of AI-driven emotional support systems.

Bobo Li, Rui Wu, Zibo Ji et al.

Large Language Model agents have rapidly evolved from static text generators into dynamic systems capable of executing complex autonomous workflows. To enhance reliability, multi-agent frameworks assigning specialized roles are increasingly adopted to enable self-reflection and mutual auditing. While such role-playing effectively leverages domain expert knowledge, we find it simultaneously induces a human-like cognitive bias known as Actor-Observer Asymmetry (AOA). Specifically, an agent acting as an actor (during self-reflection) tends to attribute failures to external factors, whereas an observer (during mutual auditing) attributes the same errors to internal faults. We quantify this using our new Ambiguous Failure Benchmark, which reveals that simply swapping perspectives triggers the AOA effect in over 20% of cases for most models. To tame this bias, we introduce ReTAS (Reasoning via Thesis-Antithesis-Synthesis), a model trained through dialectical alignment to enforce perspective-invariant reasoning. By integrating dialectical chain-of-thought with Group Relative Policy Optimization, ReTAS guides agents to synthesize conflicting viewpoints into an objective consensus. Experiments demonstrate that ReTAS effectively mitigates attribution inconsistency and significantly improves fault resolution rates in ambiguous scenarios.

Zehong Yan, Peng Qi, Wynne Hsu et al.

Multimodal misinformation, encompassing textual, visual, and cross-modal distortions, poses an increasing societal threat that is amplified by generative AI. Existing methods typically focus on a single type of distortion and struggle to generalize to unseen scenarios. In this work, we observe that different distortion types share common reasoning capabilities while also requiring task-specific skills. We hypothesize that joint training across distortion types facilitates knowledge sharing and enhances the model's ability to generalize. To this end, we introduce TRUST-VL, a unified and explainable vision-language model for general multimodal misinformation detection. TRUST-VL incorporates a novel Question-Aware Visual Amplifier module, designed to extract task-specific visual features. To support training, we also construct TRUST-Instruct, a large-scale instruction dataset containing 198K samples featuring structured reasoning chains aligned with human fact-checking workflows. Extensive experiments on both in-domain and zero-shot benchmarks demonstrate that TRUST-VL achieves state-of-the-art performance, while also offering strong generalization and interpretability.

Bobo Li, Yuheng Wang, Hao Fei et al.

Online form filling is a common yet labor-intensive task involving extensive keyboard and mouse interactions. Despite the long-standing vision of automating this process with "one click", existing tools remain largely rule-based and lack generalizable, generative capabilities. Recent advances in Multimodal Large Language Models (MLLMs) have enabled promising agents for GUI-related tasks in general-purpose scenarios. However, they struggle with the unique challenges of form filling, such as flexible layouts and the difficulty of aligning textual instructions with on-screen fields. To bridge this gap, we formally define the form-filling task and propose FormFactory, an interactive benchmarking suite comprising a web-based interface, backend evaluation module, and carefully constructed dataset. Our benchmark covers diverse real-world scenarios, incorporates various field formats, and simulates high-fidelity form interactions. We conduct a comprehensive evaluation of state-of-the-art MLLMs and observe that no model surpasses 5% accuracy, underscoring the inherent difficulty of the task. These findings also reveal significant limitations in current models' visual layout reasoning and field-value alignment abilities. We hope our benchmark can serve as a stepping stone for further research into robust, practical form-filling agents.

Anab Maulana Barik, Wynne Hsu, Mong Li Lee

Temporal claims, often riddled with inaccuracies, are a significant challenge in the digital misinformation landscape. Fact-checking systems that can accurately verify such claims are crucial for combating misinformation. Current systems struggle with the complexities of evaluating the accuracy of these claims, especially when they include multiple, overlapping, or recurring events. We introduce a novel timeline-based fact verification framework that identify events from both claim and evidence and organize them into their respective chronological timelines. The framework systematically examines the relationships between the events in both claim and evidence to predict the veracity of each claim event and their chronological accuracy. This allows us to accurately determine the overall veracity of the claim. We also introduce a new dataset of complex temporal claims involving timeline-based reasoning for the training and evaluation of our proposed framework. Experimental results demonstrate the effectiveness of our approach in handling the intricacies of temporal claim verification.

Meng Luo, Bobo Li, Shanqing Xu et al.

Despite rapid progress in multimodal large language models (MLLMs), their capability for deep emotional understanding remains limited. We argue that genuine affective intelligence requires explicit modeling of Theory of Mind (ToM), the cognitive substrate from which emotions arise. To this end, we introduce HitEmotion, a ToM-grounded hierarchical benchmark that diagnoses capability breakpoints across increasing levels of cognitive depth. Second, we propose a ToM-guided reasoning chain that tracks mental states and calibrates cross-modal evidence to achieve faithful emotional reasoning. We further introduce TMPO, a reinforcement learning method that uses intermediate mental states as process-level supervision to guide and strengthen model reasoning. Extensive experiments show that HitEmotion exposes deep emotional reasoning deficits in state-of-the-art models, especially on cognitively demanding tasks. In evaluation, the ToM-guided reasoning chain and TMPO improve end-task accuracy and yield more faithful, more coherent rationales. In conclusion, our work provides the research community with a practical toolkit for evaluating and enhancing the cognition-based emotional understanding capabilities of MLLMs. Our dataset and code are available at: https://HitEmotion.github.io/.

Wei Liu, Shengqiong Wu, Bobo Li et al.

In the AIGC era, generating high-quality 4D content has garnered increasing research attention. Unfortunately, current 4D synthesis research is severely constrained by the lack of large-scale 4D datasets, preventing models from adequately learning the critical spatial-temporal features necessary for high-quality 4D generation, thus hindering progress in this domain. To combat this, we propose a novel framework that transfers rich spatial priors from existing 3D diffusion models and temporal priors from video diffusion models to enhance 4D synthesis. We develop a spatial-temporal-disentangled 4D (STD-4D) Diffusion model, which synthesizes 4D-aware videos through disentangled spatial and temporal latents. To facilitate the best feature transfer, we design a novel Orthogonal Spatial-temporal Distributional Transfer (Orster) mechanism, where the spatiotemporal feature distributions are carefully modeled and injected into the STD-4D Diffusion. Furthermore, during the 4D construction, we devise a spatial-temporal-aware HexPlane (ST-HexPlane) to integrate the transferred spatiotemporal features, thereby improving 4D deformation and 4D Gaussian feature modeling. Experiments demonstrate that our method significantly outperforms existing approaches, achieving superior spatial-temporal consistency and higher-quality 4D synthesis.

Yanlin Li, Minghui Guo, Kaiwen Zhang et al.

In real-world multimodal applications, systems usually need to comprehend arbitrarily combined and interleaved multimodal inputs from users, while also generating outputs in any interleaved multimedia form. This capability defines the goal of any-to-any interleaved multimodal learning under a unified paradigm of understanding and generation, posing new challenges and opportunities for advancing Multimodal Large Language Models (MLLMs). To foster and benchmark this capability, this paper introduces the UniM benchmark, the first Unified Any-to-Any Interleaved Multimodal dataset. UniM contains 31K high-quality instances across 30 domains and 7 representative modalities: text, image, audio, video, document, code, and 3D, each requiring multiple intertwined reasoning and generation capabilities. We further introduce the UniM Evaluation Suite, which assesses models along three dimensions: Semantic Correctness & Generation Quality, Response Structure Integrity, and Interleaved Coherence. In addition, we propose UniMA, an agentic baseline model equipped with traceable reasoning for structured interleaved generation. Comprehensive experiments demonstrate the difficulty of UniM and highlight key challenges and directions for advancing unified any-to-any multimodal intelligence. The project page is https://any2any-mllm.github.io/unim.

Yingnan Liu, Rui Qiao, Mong Li Lee et al.

Test-time adaptation aims to improve model robustness under distribution shifts by adapting models with access to unlabeled target samples. A primary cause of performance degradation under such shifts is the model's reliance on features that lack a direct causal relationship with the prediction target. We introduce Test-time Adaptation by Causal Trimming (TACT), a method that identifies and removes non-causal components from representations for test distributions. TACT applies data augmentations that preserve causal features while varying non-causal ones. By analyzing the changes in the representations using Principal Component Analysis, TACT identifies the highest variance directions associated with non-causal features. It trims the representations by removing their projections on the identified directions, and uses the trimmed representations for the predictions. During adaptation, TACT continuously tracks and refines these directions to get a better estimate of non-causal features. We theoretically analyze the effectiveness of this approach and empirically validate TACT on real-world out-of-distribution benchmarks. TACT consistently outperforms state-of-the-art methods by a significant margin.

Jundong Xu, Hao Fei, Yuhui Zhang et al.

Multimodal symbolic logical reasoning, which aims to deduce new facts from multimodal input via formal logic, is critical in high-stakes applications such as autonomous driving and medical diagnosis, as its rigorous, deterministic reasoning helps prevent serious consequences. To evaluate such capabilities of current state-of-the-art vision language models (VLMs), we introduce the first benchmark MuSLR for multimodal symbolic logical reasoning grounded in formal logical rules. MuSLR comprises 1,093 instances across 7 domains, including 35 atomic symbolic logic and 976 logical combinations, with reasoning depths ranging from 2 to 9. We evaluate 7 state-of-the-art VLMs on MuSLR and find that they all struggle with multimodal symbolic reasoning, with the best model, GPT-4.1, achieving only 46.8%. Thus, we propose LogiCAM, a modular framework that applies formal logical rules to multimodal inputs, boosting GPT-4.1's Chain-of-Thought performance by 14.13%, and delivering even larger gains on complex logics such as first-order logic. We also conduct a comprehensive error analysis, showing that around 70% of failures stem from logical misalignment between modalities, offering key insights to guide future improvements. All data and code are publicly available at https://llm-symbol.github.io/MuSLR.

Lanhu Wu, Zilin Gao, Hao Fei et al.

RGB-D salient object detection (SOD) aims to identify the most conspicuous objects in a scene with the incorporation of depth cues. Existing methods mainly rely on CNNs, limited by the local receptive fields, or Vision Transformers that suffer from the cost of quadratic complexity, posing a challenge in balancing performance and computational efficiency. Recently, state space models (SSM), Mamba, have shown great potential for modeling long-range dependency with linear complexity. However, directly applying SSM to RGB-D SOD may lead to deficient local semantics as well as the inadequate cross-modality fusion. To address these issues, we propose a Local Emphatic and Adaptive Fusion state space model (LEAF-Mamba) that contains two novel components: 1) a local emphatic state space module (LE-SSM) to capture multi-scale local dependencies for both modalities. 2) an SSM-based adaptive fusion module (AFM) for complementary cross-modality interaction and reliable cross-modality integration. Extensive experiments demonstrate that the LEAF-Mamba consistently outperforms 16 state-of-the-art RGB-D SOD methods in both efficacy and efficiency. Moreover, our method can achieve excellent performance on the RGB-T SOD task, proving a powerful generalization ability.

Zehong Yan, Peng Qi, Wynne Hsu et al.

While large generative artificial intelligence (GenAI) models have achieved significant success, they also raise growing concerns about online information security due to their potential misuse for generating deceptive content. Out-of-context (OOC) multimodal misinformation detection, which often retrieves Web evidence to identify the repurposing of images in false contexts, faces the issue of reasoning over GenAI-polluted evidence to derive accurate predictions. Existing works simulate GenAI-powered pollution at the claim level with stylistic rewriting to conceal linguistic cues, and ignore evidence-level pollution for such information-seeking applications. In this work, we investigate how polluted evidence affects the performance of existing OOC detectors, revealing a performance degradation of more than 9 percentage points. We propose two strategies, cross-modal evidence reranking and cross-modal claim-evidence reasoning, to address the challenges posed by polluted evidence. Extensive experiments on two benchmark datasets show that these strategies can effectively enhance the robustness of existing out-of-context detectors amidst polluted evidence.

Jay Nandy, Wynne Hsu, Mong Li Lee

Deep learning-based models are developed to automatically detect if a retina image is `referable' in diabetic retinopathy (DR) screening. However, their classification accuracy degrades as the input images distributionally shift from their training distribution. Further, even if the input is not a retina image, a standard DR classifier produces a high confident prediction that the image is `referable'. Our paper presents a Dirichlet Prior Network-based framework to address this issue. It utilizes an out-of-distribution (OOD) detector model and a DR classification model to improve generalizability by identifying OOD images. Experiments on real-world datasets indicate that the proposed framework can eliminate the unknown non-retina images and identify the distributionally shifted retina images for human intervention.

Feng Xie, Han Yuan, Yilin Ning et al.

Objective: Temporal electronic health records (EHRs) can be a wealth of information for secondary uses, such as clinical events prediction or chronic disease management. However, challenges exist for temporal data representation. We therefore sought to identify these challenges and evaluate novel methodologies for addressing them through a systematic examination of deep learning solutions. Methods: We searched five databases (PubMed, EMBASE, the Institute of Electrical and Electronics Engineers [IEEE] Xplore Digital Library, the Association for Computing Machinery [ACM] digital library, and Web of Science) complemented with hand-searching in several prestigious computer science conference proceedings. We sought articles that reported deep learning methodologies on temporal data representation in structured EHR data from January 1, 2010, to August 30, 2020. We summarized and analyzed the selected articles from three perspectives: nature of time series, methodology, and model implementation. Results: We included 98 articles related to temporal data representation using deep learning. Four major challenges were identified, including data irregularity, data heterogeneity, data sparsity, and model opacity. We then studied how deep learning techniques were applied to address these challenges. Finally, we discuss some open challenges arising from deep learning. Conclusion: Temporal EHR data present several major challenges for clinical prediction modeling and data utilization. To some extent, current deep learning solutions can address these challenges. Future studies can consider designing comprehensive and integrated solutions. Moreover, researchers should incorporate additional clinical domain knowledge into study designs and enhance the interpretability of the model to facilitate its implementation in clinical practice.

Sandareka Wickramanayake, Wynne Hsu, Mong Li Lee

Despite the remarkable performance, Deep Neural Networks (DNNs) behave as black-boxes hindering user trust in Artificial Intelligence (AI) systems. Research on opening black-box DNN can be broadly categorized into post-hoc methods and inherently interpretable DNNs. While many surveys have been conducted on post-hoc interpretation methods, little effort is devoted to inherently interpretable DNNs. This paper provides a review of existing methods to develop DNNs with intrinsic interpretability, with a focus on Convolutional Neural Networks (CNNs). The aim is to understand the current progress towards fully interpretable DNNs that can cater to different interpretation requirements. Finally, we identify gaps in current work and suggest potential research directions.

Jay Nandy, Sudipan Saha, Wynne Hsu et al.

The current state-of-the-art defense methods against adversarial examples typically focus on improving either empirical or certified robustness. Among them, adversarially trained (AT) models produce empirical state-of-the-art defense against adversarial examples without providing any robustness guarantees for large classifiers or higher-dimensional inputs. In contrast, existing randomized smoothing based models achieve state-of-the-art certified robustness while significantly degrading the empirical robustness against adversarial examples. In this paper, we propose a novel method, called \emph{Certification through Adaptation}, that transforms an AT model into a randomized smoothing classifier during inference to provide certified robustness for $\ell_2$ norm without affecting their empirical robustness against adversarial attacks. We also propose \emph{Auto-Noise} technique that efficiently approximates the appropriate noise levels to flexibly certify the test examples using randomized smoothing technique. Our proposed \emph{Certification through Adaptation} with \emph{Auto-Noise} technique achieves an \textit{average certified radius (ACR) scores} up to $1.102$ and $1.148$ respectively for CIFAR-10 and ImageNet datasets using AT models without affecting their empirical robustness or benign accuracy. Therefore, our paper is a step towards bridging the gap between the empirical and certified robustness against adversarial examples by achieving both using the same classifier.

Sandareka Wickramanayake, Wynne Hsu, Mong Li Lee

Learning concepts that are consistent with human perception is important for Deep Neural Networks to win end-user trust. Post-hoc interpretation methods lack transparency in the feature representations learned by the models. This work proposes a guided learning approach with an additional concept layer in a CNN- based architecture to learn the associations between visual features and word phrases. We design an objective function that optimizes both prediction accuracy and semantics of the learned feature representations. Experiment results demonstrate that the proposed model can learn concepts that are consistent with human perception and their corresponding contributions to the model decision without compromising accuracy. Further, these learned concepts are transferable to new classes of objects that have similar concepts.

Jay Nandy, Wynne Hsu, Mong Li Lee

Among existing uncertainty estimation approaches, Dirichlet Prior Network (DPN) distinctly models different predictive uncertainty types. However, for in-domain examples with high data uncertainties among multiple classes, even a DPN model often produces indistinguishable representations from the out-of-distribution (OOD) examples, compromising their OOD detection performance. We address this shortcoming by proposing a novel loss function for DPN to maximize the \textit{representation gap} between in-domain and OOD examples. Experimental results demonstrate that our proposed approach consistently improves OOD detection performance.

Suman Bhoi, Lee Mong Li, Wynne Hsu

The broad adoption of Electronic Health Records (EHR) has led to vast amounts of data being accumulated on a patient's history, diagnosis, prescriptions, and lab tests. Advances in recommender technologies have the potential to utilize this information to help doctors personalize the prescribed medications. In this work, we design a two-stage attention-based personalized medication recommender system called PREMIER which incorporates information from the EHR to suggest a set of medications. Our system takes into account the interactions among drugs in order to minimize the adverse effects for the patient. We utilize the various attention weights in the system to compute the contributions from the information sources for the recommended medications. Experiment results on MIMIC-III and a proprietary outpatient dataset show that PREMIER outperforms state-of-the-art medication recommendation systems while achieving the best tradeoff between accuracy and drug-drug interaction. Two case studies are also presented demonstrating that the justifications provided by PREMIER are appropriate and aligned to clinical practices.

Jay Nandy, Wynne Hsu, Mong Li Lee

Existing defenses against adversarial attacks are typically tailored to a specific perturbation type. Using adversarial training to defend against multiple types of perturbation requires expensive adversarial examples from different perturbation types at each training step. In contrast, manifold-based defense incorporates a generative network to project an input sample onto the clean data manifold. This approach eliminates the need to generate expensive adversarial examples while achieving robustness against multiple perturbation types. However, the success of this approach relies on whether the generative network can capture the complete clean data manifold, which remains an open problem for complex input domain. In this work, we devise an approximate manifold defense mechanism, called RBF-CNN, for image classification. Instead of capturing the complete data manifold, we use an RBF layer to learn the density of small image patches. RBF-CNN also utilizes a reconstruction layer that mitigates any minor adversarial perturbations. Further, incorporating our proposed reconstruction process for training improves the adversarial robustness of our RBF-CNN models. Experiment results on MNIST and CIFAR-10 datasets indicate that RBF-CNN offers robustness for multiple perturbations without the need for expensive adversarial training.

Jay Nandy, Wynne Hsu, Mong Li Lee

Gaussian distributions are commonly used as a key building block in many generative models. However, their applicability has not been well explored in deep networks. In this paper, we propose a novel deep generative model named as Normal Similarity Network (NSN) where the layers are constructed with Gaussian-style filters. NSN is trained with a layer-wise non-parametric density estimation algorithm that iteratively down-samples the training images and captures the density of the down-sampled training images in the final layer. Additionally, we propose NSN-Gen for generating new samples from noise vectors by iteratively reconstructing feature maps in the hidden layers of NSN. Our experiments suggest encouraging results of the proposed model for a wide range of computer vision applications including image generation, styling and reconstruction from occluded images.

Lahari Poddar, Wynne Hsu, Mong Li Lee

User opinions expressed in the form of ratings can influence an individual's view of an item. However, the true quality of an item is often obfuscated by user biases, and it is not obvious from the observed ratings the importance different users place on different aspects of an item. We propose a probabilistic modeling of the observed aspect ratings to infer (i) each user's aspect bias and (ii) latent intrinsic quality of an item. We model multi-aspect ratings as ordered discrete data and encode the dependency between different aspects by using a latent Gaussian structure. We handle the Gaussian-Categorical non-conjugacy using a stick-breaking formulation coupled with Pólya-Gamma auxiliary variable augmentation for a simple, fully Bayesian inference. On two real world datasets, we demonstrate the predictive ability of our model and its effectiveness in learning explainable user biases to provide insights towards a more reliable product quality estimation.