Junchi Yu

Junchi Yu, Jian Liang, Ran He

Out-of-distribution (OOD) generalization is an important issue for Graph Neural Networks (GNNs). Recent works employ different graph editions to generate augmented environments and learn an invariant GNN for generalization. However, the label shift usually occurs in augmentation since graph structural edition inevitably alters the graph label. This brings inconsistent predictive relationships among augmented environments, which is harmful to generalization. To address this issue, we propose \textbf{LiSA}, which generates label-invariant augmentations to facilitate graph OOD generalization. Instead of resorting to graph editions, LiSA exploits \textbf{L}abel-\textbf{i}nvariant \textbf{S}ubgraphs of the training graphs to construct \textbf{A}ugmented environments. Specifically, LiSA first designs the variational subgraph generators to extract locally predictive patterns and construct multiple label-invariant subgraphs efficiently. Then, the subgraphs produced by different generators are collected to build different augmented environments. To promote diversity among augmented environments, LiSA further introduces a tractable energy-based regularization to enlarge pair-wise distances between the distributions of environments. In this manner, LiSA generates diverse augmented environments with a consistent predictive relationship and facilitates learning an invariant GNN. Extensive experiments on node-level and graph-level OOD benchmarks show that LiSA achieves impressive generalization performance with different GNN backbones. Code is available on \url{https://github.com/Samyu0304/LiSA}.

Kaiwei Zhang, Junchi Yu, Haichao Shi et al.

The growth in social media has exacerbated the threat of fake news to individuals and communities. This draws increasing attention to developing efficient and timely rumor detection methods. The prevailing approaches resort to graph neural networks (GNNs) to exploit the post-propagation patterns of the rumor-spreading process. However, these methods lack inherent interpretation of rumor detection due to the black-box nature of GNNs. Moreover, these methods suffer from less robust results as they employ all the propagation patterns for rumor detection. In this paper, we address the above issues with the proposed Diverse Counterfactual Evidence framework for Rumor Detection (DCE-RD). Our intuition is to exploit the diverse counterfactual evidence of an event graph to serve as multi-view interpretations, which are further aggregated for robust rumor detection results. Specifically, our method first designs a subgraph generation strategy to efficiently generate different subgraphs of the event graph. We constrain the removal of these subgraphs to cause the change in rumor detection results. Thus, these subgraphs naturally serve as counterfactual evidence for rumor detection. To achieve multi-view interpretation, we design a diversity loss inspired by Determinantal Point Processes (DPP) to encourage diversity among the counterfactual evidence. A GNN-based rumor detection model further aggregates the diverse counterfactual evidence discovered by the proposed DCE-RD to achieve interpretable and robust rumor detection results. Extensive experiments on two real-world datasets show the superior performance of our method. Our code is available at https://github.com/Vicinity111/DCE-RD.

Xiaoxi Luo, Zachary Shinnick, Niclas Griesshaber et al.

We introduce TypewriterLM, a 7.24B History language model (LM) trained exclusively on English text predating 1913. Developing History LMs requires addressing challenges in data quality and availability, preventing temporal leakage, designing temporally consistent post-training pipelines, and constructing reliable evaluations. To address these issues, we construct TypewriterCorpus, a 54B-token historical corpus collected from diverse archival and linguistically annotated sources with extensive data cleaning and leakage mitigation procedures. Furthermore, we introduce lexically grounded instructing tuning, a post-training framework that constraints responses to remain directly grounded in historical source documents. Using this framework we construct two historical instruction tuning datasets: History-LIMA and History-SelfInstruct. To evaluate capability and temporal consistency, we introduce History-Event, a benchmark suite for evaluating competence, temporal grounding and data leakage. We release TypewriterLM and all associated resources to support future research on historical language models.

Bingzhe Wu, Jintang Li, Junchi Yu et al.

Deep graph learning has achieved remarkable progresses in both business and scientific areas ranging from finance and e-commerce, to drug and advanced material discovery. Despite these progresses, how to ensure various deep graph learning algorithms behave in a socially responsible manner and meet regulatory compliance requirements becomes an emerging problem, especially in risk-sensitive domains. Trustworthy graph learning (TwGL) aims to solve the above problems from a technical viewpoint. In contrast to conventional graph learning research which mainly cares about model performance, TwGL considers various reliability and safety aspects of the graph learning framework including but not limited to robustness, explainability, and privacy. In this survey, we provide a comprehensive review of recent leading approaches in the TwGL field from three dimensions, namely, reliability, explainability, and privacy protection. We give a general categorization for existing work and review typical work for each category. To give further insights for TwGL research, we provide a unified view to inspect previous works and build the connection between them. We also point out some important open problems remaining to be solved in the future developments of TwGL.

Fabian Degen, Oishi Deb, Jindong Gu et al.

Planning records define restrictions over geographic areas, but their source documents often provide only indirect spatial evidence rather than machine-readable boundaries. We introduce Plan2Map, a 208-case multimodal benchmark for document-grounded geospatial boundary reconstruction from UK planning records. Given only a source planning document, systems must reconstruct a valid geospatial boundary from notice text, schedules, map plates, map labels, and boundary annotations; the reference GeoJSON is held out for scoring. We propose GeoPlanAgent, a document-grounded, geospatial-tool-in-the-loop system that decomposes the task into evidence extraction, localisation, map registration, boundary segmentation, projection, and verification. On Plan2Map, GeoPlanAgent achieves 0.736 mean IoU and 0.904 median IoU, with 67.8\% of predictions at or above 0.8 IoU, substantially outperforming direct VLM-to-GeoJSON baselines. Diagnostic analysis shows that direct VLM prediction remains unreliable, while remaining errors are concentrated in localisation and map registration, and supervised boundary segmentation substantially improves pixel-level mask quality. Plan2Map provides a concrete testbed for multimodal geospatial reconstruction from public planning records. Project page: https://odeb1.github.io/Plan2Map_Project_Page/.

Bohang Sun, Max Zhu, Francesco Caso et al.

Diffusion large language models promise faster generation by refining many token positions in parallel, but this parallelism introduces a hidden control problem: which proposed tokens should be transferred into the partially decoded sequence at each step? We refer to this decision as token commitment. Existing frozen-generator decoders largely rely on hand-designed confidence rules or block-specific acceptance filters. We argue that token commitment can instead be learned as a reusable trace-state policy. We introduce TraceLock, a lightweight plug-in controller that instantiates this policy for a frozen diffusion language model. Since oracle commitment times are unavailable, TraceLock derives self-supervision from future stability: at decoding step t, a proposed token for position i is labeled stable if it matches the final token at position i after the full decoding trace completes. The controller scores variable-length trace states and decides which active token proposals should be committed to the partially decoded sequence. Once trained for a given frozen backbone, the controller can be deployed across local-window widths, generation lengths, and step budgets without retraining or per-setting calibration. Experiments on question answering, mathematical reasoning, and code generation show that TraceLock improves the quality-step tradeoff over heuristic and learned baselines, with particularly stable behavior under cross-setting deployment. Diagnostic analyses show that its decisions are not reducible to scalar confidence, suggesting that frozen diffusion language models expose a learnable space of commitment trajectories beyond confidence-based decoding. Code is available at https://github.com/BobSun98/TraceLock.

Puning Yang, Junchi Yu, Qizhou Wang et al.

Mitigating sensitive and harmful outputs is fundamental to ensuring safe deployment of LLMs. Existing approaches typically follow two paradigms: Knowledge Deletion (KD), which erases undesirable information during training, and Distinguishable Refusal (DR), which steers models away from using sensitive knowledge during inference. Despite rapid progress, KD-based unlearning struggles with biased deletion due to suppressing specific token sequences as a substitute for complete knowledge removal, whereas DR-based unlearning risks the re-emergence of harmful knowledge because the underlying knowledge remains intact. To address these issues, we propose Distinguishable Deletion ($\mathrm{D^2}$), a paradigm that restricts the response distribution in the latent representation rather than specific tokens to erase undesirable knowledge, while distinguishing it from retained knowledge, enabling a refusal mechanism to handle unlearned inputs safely and coherently. To implement $\mathrm{D^2}$, we introduce an energy index that quantifies the presence of knowledge and the separation between unlearned and retained content. Mathematical and empirical analyses show that energy is both accurate and efficient, enabling Energy-based Unlearning Alignment (EUA) to enforce energy-boundary unlearning during training and apply an energy-based refusal mechanism at inference. Extensive experiments demonstrate that EUA significantly outperforms previous methods, indicating the superiority of $\mathrm{D^2}$. Our code is available at https://github.com/Puning97/EUA-for-LLM-Unlearning.

Junchi Yu, Jian Liang, Ran He

This paper focuses on out-of-distribution generalization on graphs where performance drops due to the unseen distribution shift. Previous graph domain generalization works always resort to learning an invariant predictor among different source domains. However, they assume sufficient source domains are available during training, posing huge challenges for realistic applications. By contrast, we propose a new graph domain generalization framework, dubbed as DPS, by constructing multiple populations from the source domains. Specifically, DPS aims to discover multiple \textbf{D}iverse and \textbf{P}redictable \textbf{S}ubgraphs with a set of generators, namely, subgraphs are different from each other but all the them share the same semantics with the input graph. These generated source domains are exploited to learn an \textit{equi-predictive} graph neural network (GNN) across domains, which is expected to generalize well to unseen target domains. Generally, DPS is model-agnostic that can be incorporated with various GNN backbones. Extensive experiments on both node-level and graph-level benchmarks shows that the proposed DPS achieves impressive performance for various graph domain generalization tasks.

Junchi Yu, Ran He, Rex Ying

Large Language Models (LLMs) have achieved remarkable success in reasoning tasks with the development of prompting methods. However, existing prompting approaches cannot reuse insights of solving similar problems and suffer from accumulated errors in multi-step reasoning, since they prompt LLMs to reason \textit{from scratch}. To address these issues, we propose \textbf{\textit{Thought Propagation} (TP)}, which explores the analogous problems and leverages their solutions to enhance the complex reasoning ability of LLMs. These analogous problems are related to the input one, with reusable solutions and problem-solving strategies. Thus, it is promising to propagate insights of solving previous analogous problems to inspire new problem-solving. To achieve this, TP first prompts LLMs to propose and solve a set of analogous problems that are related to the input one. Then, TP reuses the results of analogous problems to directly yield a new solution or derive a knowledge-intensive plan for execution to amend the initial solution obtained from scratch. TP is compatible with existing prompting approaches, allowing plug-and-play generalization and enhancement in a wide range of tasks without much labor in task-specific prompt engineering. Experiments across three challenging tasks demonstrate TP enjoys a substantial improvement over the baselines by an average of 12\% absolute increase in finding the optimal solutions in Shortest-path Reasoning, 13\% improvement of human preference in Creative Writing, and 15\% enhancement in the task completion rate of LLM-Agent Planning.

Jie Cao, Mandi Luo, Junchi Yu et al.

Generative Adversarial Networks (GANs) typically suffer from overfitting when limited training data is available. To facilitate GAN training, current methods propose to use data-specific augmentation techniques. Despite the effectiveness, it is difficult for these methods to scale to practical applications. In this work, we present ScoreMix, a novel and scalable data augmentation approach for various image synthesis tasks. We first produce augmented samples using the convex combinations of the real samples. Then, we optimize the augmented samples by minimizing the norms of the data scores, i.e., the gradients of the log-density functions. This procedure enforces the augmented samples close to the data manifold. To estimate the scores, we train a deep estimation network with multi-scale score matching. For different image synthesis tasks, we train the score estimation network using different data. We do not require the tuning of the hyperparameters or modifications to the network architecture. The ScoreMix method effectively increases the diversity of data and reduces the overfitting problem. Moreover, it can be easily incorporated into existing GAN models with minor modifications. Experimental results on numerous tasks demonstrate that GAN models equipped with the ScoreMix method achieve significant improvements.

Aoxi Liu, Yupeng Chen, James Oldfield et al.

Despite the emergence of diffusion large language models (D-LLMs) as an alternative to autoregressive large language models (AR-LLMs), safety monitoring for D-LLMs remains largely unexplored. Unlike AR-LLMs, D-LLMs generate text through a multi-step denoising process, exposing intermediate hidden representations that may contain safety-relevant information unavailable in standard single-step monitoring setups. Motivated by the suitability of lightweight probes for always-on monitoring, we analyze which trajectory-level signals best indicate when such probes are likely to struggle. We find that the most informative signal is safety hesitation: intermediate hidden states repeatedly falling within a small margin of the probe's decision boundary. The number of such hesitation steps in D-LLM's trajectory predicts probe failure effectively, providing a proxy of sample difficulty. Building on this analysis, we propose $D^2$-Monitor, a bi-level safety monitor for D-LLMs. $D^2$-Monitor adopts a lightweight probe as an always-on monitor to jointly estimate hesitation and perform base classification. When the hesitation level exceeds a threshold, a more expressive but computationally heavier probe is activated. This dynamic routing mechanism allocates monitoring resources efficiently at test time. Evaluated on 3 datasets (WildguardMix, ToxicChat, OpenAI-Moderation) across 4 D-LLMs, $D^2$-Monitor achieves state-of-the-art performance with a compact parameter footprint ($\leq$ 0.85M parameters), and exhibits the best trade-off between effectiveness and efficiency relative to 8 baselines.

Sean Wu, Pan Lu, Yupeng Chen et al.

Artificial intelligence (AI) is increasingly embedded in scientific discovery, yet whether it can anticipate scientific progress remains unclear. To study this question, we introduce a temporally grounded evaluation framework for forecasting scientific progress under controlled knowledge constraints. We present CUSP (Cutoff-conditioned Unseen Scientific Progress), a multi-disciplinary and event-level benchmark that evaluates scientific forecasting in AI systems through feasibility assessment, mechanistic reasoning, generative solution design, and temporal prediction. Across 4,760 scientific events, we observe systematic and domain-dependent limitations in current frontier models. While models can identify plausible research directions from competing candidates, they fail to reliably predict whether scientific advances will be realized and systematically misestimate when they will occur. Performance is highly heterogeneous across domains, with the timing of AI progress more predictable than advances in biology, chemistry, and physics. Performance is largely insensitive to whether events occur before or after the training cutoff, suggesting these limitations cannot be explained solely by knowledge exposure in training data. Under controlled information access, additional pre-cutoff knowledge improves performance but does not close the gap to full-information settings, which becomes more pronounced for high-citation advances. Models also exhibit systematic overconfidence and strong response biases, indicating unreliable uncertainty estimation. Taken together, current AI systems fall short as predictive tools for scientific progress. Access to prior knowledge does not translate into reliable forecasting, and performance benefits more from post-event information than from forward-looking prediction.

Jialin Chen, Kenza Amara, Junchi Yu et al.

Graph Neural Networks (GNNs) achieve state-of-the-art performance in various graph-related tasks. However, the black-box nature often limits their interpretability and trustworthiness. Numerous explainability methods have been proposed to uncover the decision-making logic of GNNs, by generating underlying explanatory substructures. In this paper, we conduct a comprehensive review of the existing explanation methods for GNNs from the perspective of graph generation. Specifically, we propose a unified optimization objective for generative explanation methods, comprising two sub-objectives: Attribution and Information constraints. We further demonstrate their specific manifestations in various generative model architectures and different explanation scenarios. With the unified objective of the explanation problem, we reveal the shared characteristics and distinctions among current methods, laying the foundation for future methodological advancements. Empirical results demonstrate the advantages and limitations of different explainability approaches in terms of explanation performance, efficiency, and generalizability.

Zeyuan He, Yupeng Chen, Lang Lin et al.

Diffusion large language models (D-LLMs) offer an alternative to autoregressive LLMs (AR-LLMs) and have demonstrated advantages in generation efficiency. Beyond the utility benefits, we argue that D-LLMs exhibit a previously underexplored safety blessing: their diffusion-style generation confers intrinsic robustness against jailbreak attacks originally designed for AR-LLMs. In this work, we provide an initial analysis of the underlying mechanism, showing that the diffusion trajectory induces a stepwise reduction effect that progressively suppresses unsafe generations. This robustness, however, is not absolute. We identify a simple yet effective failure mode, termed context nesting, where harmful requests are embedded within structured benign contexts, effectively bypassing the stepwise reduction mechanism. Empirically, we show that this simple strategy is sufficient to bypass D-LLMs' safety blessing, achieving state-of-the-art attack success rates across models and benchmarks. Most notably, it enables the first successful jailbreak of Gemini Diffusion, to our knowledge, exposing a critical vulnerability in commercial D-LLMs. Together, our results characterize both the origins and the limits of D-LLMs' safety blessing, constituting an early-stage red-teaming of D-LLMs.

Yupeng Chen, Junchi Yu, Aoxi Liu et al.

Recent advances in end-to-end trained omni-models have significantly improved multimodal understanding. At the same time, safety red-teaming has expanded beyond text to encompass audio-based jailbreak attacks. However, an important bridge between textual and audio jailbreaks remains underexplored. In this work, we study the cross-modality transfer of jailbreak attacks from text to audio, motivated by the semantic similarity between the two modalities and the maturity of textual jailbreak methods. We first analyze the connection between modality alignment and cross-modality jailbreak transfer, showing that strong alignment can inadvertently propagate textual vulnerabilities to the audio modality, which we term the alignment curse. Guided by this analysis, we conduct an empirical evaluation of textual jailbreaks, text-transferred audio jailbreaks, and existing audio-based jailbreaks on recent omni-models. Our results show that text-transferred audio jailbreaks perform comparably to, and often better than, audio-based jailbreaks, establishing them as simple yet powerful baselines for future audio red-teaming. We further demonstrate strong cross-model transferability and show that text-transferred audio attacks remain effective even under a stricter audio-only access threat model.

Pengze Li, Jiaquan Zhang, Yunbo Long et al.

Recovering analytical solutions of physical fields from visual observations is a fundamental yet underexplored capability for AI-assisted scientific reasoning. We study visual-to-symbolic analytical solution inference (ViSA) for two-dimensional linear steady-state fields: given field visualizations (and first-order derivatives) plus minimal auxiliary metadata, the model must output a single executable SymPy expression with fully instantiated numeric constants. We introduce ViSA-R2 and align it with a self-verifying, solution-centric chain-of-thought pipeline that follows a physicist-like pathway: structural pattern recognition solution-family (ansatz) hypothesis parameter derivation consistency verification. We also release ViSA-Bench, a VLM-ready synthetic benchmark covering 30 linear steady-state scenarios with verifiable analytical/symbolic annotations, and evaluate predictions by numerical accuracy, expression-structure similarity, and character-level accuracy. Using an 8B open-weight Qwen3-VL backbone, ViSA-R2 outperforms strong open-source baselines and the evaluated closed-source frontier VLMs under a standardized protocol.

Shenxu Chang, Junchi Yu, Weixing Wang et al.

Diffusion large language models (D-LLMs) have recently emerged as a promising alternative to auto-regressive LLMs (AR-LLMs). However, the hallucination problem in D-LLMs remains underexplored, limiting their reliability in real-world applications. Existing hallucination detection methods are designed for AR-LLMs and rely on signals from single-step generation, making them ill-suited for D-LLMs where hallucination signals often emerge throughout the multi-step denoising process. To bridge this gap, we propose TraceDet, a novel framework that explicitly leverages the intermediate denoising steps of D-LLMs for hallucination detection. TraceDet models the denoising process as an action trace, with each action defined as the model's prediction over the cleaned response, conditioned on the previous intermediate output. By identifying the sub-trace that is maximally informative to the hallucinated responses, TraceDet leverages the key hallucination signals in the multi-step denoising process of D-LLMs for hallucination detection. Extensive experiments on various open source D-LLMs demonstrate that TraceDet consistently improves hallucination detection, achieving an average gain in AUROC of 15.2% compared to baselines.

Junchi Yu, Jie Cao, Ran He

Subgraph recognition aims at discovering a compressed substructure of a graph that is most informative to the graph property. It can be formulated by optimizing Graph Information Bottleneck (GIB) with a mutual information estimator. However, GIB suffers from training instability and degenerated results due to its intrinsic optimization process. To tackle these issues, we reformulate the subgraph recognition problem into two steps: graph perturbation and subgraph selection, leading to a novel Variational Graph Information Bottleneck (VGIB) framework. VGIB first employs the noise injection to modulate the information flow from the input graph to the perturbed graph. Then, the perturbed graph is encouraged to be informative to the graph property. VGIB further obtains the desired subgraph by filtering out the noise in the perturbed graph. With the customized noise prior for each input, the VGIB objective is endowed with a tractable variational upper bound, leading to a superior empirical performance as well as theoretical properties. Extensive experiments on graph interpretation, explainability of Graph Neural Networks, and graph classification show that VGIB finds better subgraphs than existing methods. Code is avaliable at https://github.com/Samyu0304/VGIB

Haiyuan Wan, Chen Yang, Junchi Yu et al.

Deep research agents have attracted growing attention for their potential to orchestrate multi-stage research workflows, spanning literature synthesis, methodological design, and empirical verification. Despite these strides, evaluating their research capability faithfully is rather challenging due to the difficulty of collecting frontier research questions that genuinely capture researchers' attention and intellectual curiosity. To address this gap, we introduce DeepResearch Arena, a benchmark grounded in academic seminars that capture rich expert discourse and interaction, better reflecting real-world research environments and reducing the risk of data leakage. To automatically construct DeepResearch Arena, we propose a Multi-Agent Hierarchical Task Generation (MAHTG) system that extracts research-worthy inspirations from seminar transcripts. The MAHTG system further translates research-worthy inspirations into high-quality research tasks, ensuring the traceability of research task formulation while filtering noise. With the MAHTG system, we curate DeepResearch Arena with over 10,000 high-quality research tasks from over 200 academic seminars, spanning 12 disciplines, such as literature, history, and science. Our extensive evaluation shows that DeepResearch Arena presents substantial challenges for current state-of-the-art agents, with clear performance gaps observed across different models.

Cheng Yang, Jiaxuan Lu, Haiyuan Wan et al.

The chemical reaction recommendation is to select proper reaction condition parameters for chemical reactions, which is pivotal to accelerating chemical science. With the rapid development of large language models (LLMs), there is growing interest in leveraging their reasoning and planning capabilities for reaction condition recommendation. Despite their success, existing methods rarely explain the rationale behind the recommended reaction conditions, limiting their utility in high-stakes scientific workflows. In this work, we propose ChemMAS, a multi-agent system that reframes condition prediction as an evidence-based reasoning task. ChemMAS decomposes the task into mechanistic grounding, multi-channel recall, constraint-aware agentic debate, and rationale aggregation. Each decision is backed by interpretable justifications grounded in chemical knowledge and retrieved precedents. Experiments show that ChemMAS achieves 20-35% gains over domain-specific baselines and outperforms general-purpose LLMs by 10-15% in Top-1 accuracy, while offering falsifiable, human-trustable rationales, which establishes a new paradigm for explainable AI in scientific discovery.

Cheng Yang, Haiyuan Wan, Yiran Peng et al.

Video Models have achieved remarkable success in high-fidelity video generation with coherent motion dynamics. Analogous to the development from text generation to text-based reasoning in language modeling, the development of video models motivates us to ask: Can video models reason via video generation? Compared with the discrete text corpus, video grounds reasoning in explicit spatial layouts and temporal continuity, which serves as an ideal substrate for spatial reasoning. In this work, we explore the reasoning via video paradigm and introduce VR-Bench -- a comprehensive benchmark designed to systematically evaluate video models' reasoning capabilities. Grounded in maze-solving tasks that inherently require spatial planning and multi-step reasoning, VR-Bench contains 7,920 procedurally generated videos across five maze types and diverse visual styles. Our empirical analysis demonstrates that SFT can efficiently elicit the reasoning ability of video model. Video models exhibit stronger spatial perception during reasoning, outperforming leading VLMs and generalizing well across diverse scenarios, tasks, and levels of complexity. We further discover a test-time scaling effect, where diverse sampling during inference improves reasoning reliability by 10--20%. These findings highlight the unique potential and scalability of reasoning via video for spatial reasoning tasks.

Pengze Li, Jiaqi Liu, Junchi Yu et al.

Large language models (LLMs) are increasingly used in scientific domains. While they can produce reasoning-like content via methods such as chain-of-thought prompting, these outputs are typically unstructured and informal, obscuring whether models truly understand the fundamental reasoning paradigms that underpin scientific inference. To address this, we introduce a novel task named Latent Reasoning Chain Extraction (ARCHE), in which models must decompose complex reasoning arguments into combinations of standard reasoning paradigms in the form of a Reasoning Logic Tree (RLT). In RLT, all reasoning steps are explicitly categorized as one of three variants of Peirce's fundamental inference modes: deduction, induction, or abduction. To facilitate this task, we release ARCHE Bench, a new benchmark derived from 70 Nature Communications articles, including more than 1,900 references and 38,000 viewpoints. We propose two logic-aware evaluation metrics: Entity Coverage (EC) for content completeness and Reasoning Edge Accuracy (REA) for step-by-step logical validity. Evaluations on 10 leading LLMs on ARCHE Bench reveal that models exhibit a trade-off between REA and EC, and none are yet able to extract a complete and standard reasoning chain. These findings highlight a substantial gap between the abilities of current reasoning models and the rigor required for scientific argumentation.

Junchi Yu, Yujie Liu, Jindong Gu et al.

Retrieval-Augmented Generation (RAG) based on knowledge graphs (KGs) enhances large language models (LLMs) by providing structured and interpretable external knowledge. However, existing KG-based RAG methods struggle to retrieve accurate and diverse information from text-rich KGs for complex real-world queries. Process Reward Models (PRMs) offer a way to align the retrieval process of KG-based RAG with query-specific knowledge requirements, but they heavily rely on process-level supervision signals that are expensive and hard to obtain on KGs. To address this challenge, we propose GraphFlow, a framework that efficiently retrieves accurate and diverse knowledge required for real-world queries from text-rich KGs. GraphFlow employs a transition-based flow matching objective to jointly optimize a retrieval policy and a flow estimator. The flow estimator factorizes the reward of the retrieval outcome into the intermediate retrieval states. Such reward factorization guides the retrieval policy to retrieve candidates from KGs in proportion to their reward. This allows GraphFlow to explore high-quality regions of KGs that yield diverse and relevant results. We evaluate GraphFlow on the STaRK benchmark, which includes real-world queries from multiple domains over text-rich KGs. GraphFlow outperforms strong KG-RAG baselines, including GPT-4o, by 10% on average in hit rate and recall. It also shows strong generalization to unseen KGs, demonstrating its effectiveness and robustness.

Jialin Yu, Yuxiang Zhou, Yulan He et al.

Adapting to latent-confounded shifts remains a core challenge in modern AI. These shifts are propagated via latent variables that induce spurious, non-transportable correlations between inputs and labels. One practical failure mode arises when fine-tuning pre-trained foundation models on confounded data (e.g., where certain text tokens or image backgrounds spuriously correlate with the label), leaving models vulnerable at deployment. We frame causal fine-tuning as an identification problem and pose an explicit causal model that decomposes inputs into low-level spurious features and high-level causal representations. Under this family of models, we formalize the assumptions required for identification. Using pre-trained language models as a case study, we show how identifying and adjusting these components during causal fine-tuning enables automatic adaptation to latent-confounded shifts at test time. Experiments on semi-synthetic benchmarks derived from real-world problems demonstrate that our method outperforms black-box domain generalization baselines, illustrating the benefits of explicitly modeling causal structure.

Junchi Yu, Tingyang Xu, Ran He

As Graph Neural Networks (GNNs) are widely adopted in digital pathology, there is increasing attention to developing explanation models (explainers) of GNNs for improved transparency in clinical decisions. Existing explainers discover an explanatory subgraph relevant to the prediction. However, such a subgraph is insufficient to reveal all the critical biological substructures for the prediction because the prediction will remain unchanged after removing that subgraph. Hence, an explanatory subgraph should be not only necessary for prediction, but also sufficient to uncover the most predictive regions for the explanation. Such explanation requires a measurement of information transferred from different input subgraphs to the predictive output, which we define as information flow. In this work, we address these key challenges and propose IFEXPLAINER, which generates a necessary and sufficient explanation for GNNs. To evaluate the information flow within GNN's prediction, we first propose a novel notion of predictiveness, named $f$-information, which is directional and incorporates the realistic capacity of the GNN model. Based on it, IFEXPLAINER generates the explanatory subgraph with maximal information flow to the prediction. Meanwhile, it minimizes the information flow from the input to the predictive result after removing the explanation. Thus, the produced explanation is necessarily important to the prediction and sufficient to reveal the most crucial substructures. We evaluate IFEXPLAINER to interpret GNN's predictions on breast cancer subtyping. Experimental results on the BRACS dataset show the superior performance of the proposed method.

Junchi Yu, Tingyang Xu, Yu Rong et al.

The emergence of Graph Convolutional Network (GCN) has greatly boosted the progress of graph learning. However, two disturbing factors, noise and redundancy in graph data, and lack of interpretation for prediction results, impede further development of GCN. One solution is to recognize a predictive yet compressed subgraph to get rid of the noise and redundancy and obtain the interpretable part of the graph. This setting of subgraph is similar to the information bottleneck (IB) principle, which is less studied on graph-structured data and GCN. Inspired by the IB principle, we propose a novel subgraph information bottleneck (SIB) framework to recognize such subgraphs, named IB-subgraph. However, the intractability of mutual information and the discrete nature of graph data makes the objective of SIB notoriously hard to optimize. To this end, we introduce a bilevel optimization scheme coupled with a mutual information estimator for irregular graphs. Moreover, we propose a continuous relaxation for subgraph selection with a connectivity loss for stabilization. We further theoretically prove the error bound of our estimation scheme for mutual information and the noise-invariant nature of IB-subgraph. Extensive experiments on graph learning and large-scale point cloud tasks demonstrate the superior property of IB-subgraph.

Junchi Yu, Tingyang Xu, Yu Rong et al.

Given the input graph and its label/property, several key problems of graph learning, such as finding interpretable subgraphs, graph denoising and graph compression, can be attributed to the fundamental problem of recognizing a subgraph of the original one. This subgraph shall be as informative as possible, yet contains less redundant and noisy structure. This problem setting is closely related to the well-known information bottleneck (IB) principle, which, however, has less been studied for the irregular graph data and graph neural networks (GNNs). In this paper, we propose a framework of Graph Information Bottleneck (GIB) for the subgraph recognition problem in deep graph learning. Under this framework, one can recognize the maximally informative yet compressive subgraph, named IB-subgraph. However, the GIB objective is notoriously hard to optimize, mostly due to the intractability of the mutual information of irregular graph data and the unstable optimization process. In order to tackle these challenges, we propose: i) a GIB objective based-on a mutual information estimator for the irregular graph data; ii) a bi-level optimization scheme to maximize the GIB objective; iii) a connectivity loss to stabilize the optimization process. We evaluate the properties of the IB-subgraph in three application scenarios: improvement of graph classification, graph interpretation and graph denoising. Extensive experiments demonstrate that the information-theoretic IB-subgraph enjoys superior graph properties.

Yi Li, Huaibo Huang, Junchi Yu et al.

Face verification aims at determining whether a pair of face images belongs to the same identity. Recent studies have revealed the negative impact of facial makeup on the verification performance. With the rapid development of deep generative models, this paper proposes a semanticaware makeup cleanser (SAMC) to remove facial makeup under different poses and expressions and achieve verification via generation. The intuition lies in the fact that makeup is a combined effect of multiple cosmetics and tailored treatments should be imposed on different cosmetic regions. To this end, we present both unsupervised and supervised semantic-aware learning strategies in SAMC. At image level, an unsupervised attention module is jointly learned with the generator to locate cosmetic regions and estimate the degree. At feature level, we resort to the effort of face parsing merely in training phase and design a localized texture loss to serve complements and pursue superior synthetic quality. The experimental results on four makeuprelated datasets verify that SAMC not only produces appealing de-makeup outputs at a resolution of 256*256, but also facilitates makeup-invariant face verification through image generation.