Rui Wu

Can Jin, Jiakang Li, Rui Wu et al.

As large language models become stronger, weak supervisors may fail to provide reliable labels, preferences, or final judgments for complex outputs, limiting both weak-to-strong generalization and scalable oversight. We study a more tractable form of weak supervision: using a weak model as a critic rather than as a labeler or judge. Instead of solving the task or selecting the correct answer, the weak critic only needs to provide a non-misleading revision direction that helps the strong model better use its own knowledge. We call this setting *weak-critic strong oversight*. We first show that weak critiques can improve frozen strong models at inference time, and that critique quality is key to this improvement. We then propose progressive on-policy critique distillation (**OPCD**), which filters high-quality critiques and distills critic-guided behavior into the strong model through adaptive self-teacher signals. Experiments on reasoning and alignment benchmarks show that our method improves strong models over training epochs, suggesting an effective path for scalable oversight with weak supervision.

Jiangbei Hu, Weichao Song, Shibo Yu et al.

Underwater scene reconstruction is essential for immersive exploration of aquatic environments, yet remains challenging due to complex participating-media effects such as absorption and scattering, as well as the limited field of view (FoV) of conventional cameras. Although combining panoramic imaging with 3D Gaussian Splatting (3DGS) offers a promising direction for photorealistic underwater rendering, traditional 3DGS struggles with both spherical projection distortion and underwater medium degradation. In this paper, we propose \textbf{Underwater360}, a physics-informed omnidirectional 3DGS framework for underwater panoramic scene reconstruction. First, we introduce an Omnidirectional Gaussian Splatting module that performs ray casting directly in spherical camera space instead of relying on 2D projection approximations, thereby reducing geometric distortions under 360$^\circ$ FoV. Second, we design a physics-based appearance-medium modeling architecture with pose-conditioned appearance embeddings to explicitly decouple intrinsic scene radiance from depth-dependent backscatter and attenuation, enabling physically grounded scene appearance restoration. Finally, we establish a new panoramic underwater benchmark dataset containing both synthetic and real-world scenes. Extensive experiments demonstrate that Underwater360 achieves superior performance in underwater novel view synthesis and scene appearance restoration, delivering improved rendering quality and cross-view consistency in complex underwater environments. The code and datasets are released at https://github.com/SwcK423/Underwater360

Yifan Zhang, Rui Wu, Sergiu M. Dascalu et al.

Multivariate Time Series (MTS) forecasting involves modeling temporal dependencies within historical records. Transformers have demonstrated remarkable performance in MTS forecasting due to their capability to capture long-term dependencies. However, prior work has been confined to modeling temporal dependencies at either a fixed scale or multiple scales that exponentially increase (most with base 2). This limitation hinders their effectiveness in capturing diverse seasonalities, such as hourly and daily patterns. In this paper, we introduce a dimension invariant embedding technique that captures short-term temporal dependencies and projects MTS data into a higher-dimensional space, while preserving the dimensions of time steps and variables in MTS data. Furthermore, we present a novel Multi-scale Transformer Pyramid Network (MTPNet), specifically designed to effectively capture temporal dependencies at multiple unconstrained scales. The predictions are inferred from multi-scale latent representations obtained from transformers at various scales. Extensive experiments on nine benchmark datasets demonstrate that the proposed MTPNet outperforms recent state-of-the-art methods.

Xiaolin Zheng, Rui Wu, Zhongxuan Han et al.

Recommender systems are fundamental information filtering techniques to recommend content or items that meet users' personalities and potential needs. As a crucial solution to address the difficulty of user identification and unavailability of historical information, session-based recommender systems provide recommendation services that only rely on users' behaviors in the current session. However, most existing studies are not well-designed for modeling heterogeneous user behaviors and capturing the relationships between them in practical scenarios. To fill this gap, in this paper, we propose a novel graph-based method, namely Heterogeneous Information Crossing on Graphs (HICG). HICG utilizes multiple types of user behaviors in the sessions to construct heterogeneous graphs, and captures users' current interests with their long-term preferences by effectively crossing the heterogeneous information on the graphs. In addition, we also propose an enhanced version, named HICG-CL, which incorporates contrastive learning (CL) technique to enhance item representation ability. By utilizing the item co-occurrence relationships across different sessions, HICG-CL improves the recommendation performance of HICG. We conduct extensive experiments on three real-world recommendation datasets, and the results verify that (i) HICG achieves the state-of-the-art performance by utilizing multiple types of behaviors on the heterogeneous graph. (ii) HICG-CL further significantly improves the recommendation performance of HICG by the proposed contrastive learning module.

Ponkoj Chandra Shill, Rui Wu, Hossein Jamali et al.

This paper presents a work-in-progress on a learn-ing system that will provide robotics students with a personalized learning environment. This addresses both the scarcity of skilled robotics instructors, particularly in community colleges and the expensive demand for training equipment. The study of robotics at the college level represents a wide range of interests, experiences, and aims. This project works to provide students the flexibility to adapt their learning to their own goals and prior experience. We are developing a system to enable robotics instruction through a web-based interface that is compatible with less expensive hardware. Therefore, the free distribution of teaching materials will empower educators. This project has the potential to increase the number of robotics courses offered at both two- and four-year schools and universities. The course materials are being designed with small units and a hierarchical dependency tree in mind; students will be able to customize their course of study based on the robotics skills they have already mastered. We present an evaluation of a five module mini-course in robotics. Students indicated that they had a positive experience with the online content. They also scored the experience highly on relatedness, mastery, and autonomy perspectives, demonstrating strong motivation potential for this approach.

Songqin Nong, Jiali Zhu, Rui Wu et al.

Currently, the integration of mobile Graphical User Interfaces (GUIs) is ubiquitous in most people's daily lives. And the ongoing evolution of multimodal large-scale models, such as GPT-4v, Qwen-VL-Max, has significantly bolstered the capabilities of GUI comprehension and user action analysis, showcasing the potentiality of intelligent GUI assistants. However, current GUI Agents often need to access page layout information through calling system APIs, which may pose privacy risks. Fixing GUI (such as mobile interfaces) to a certain low resolution might result in the loss of fine-grained image details. At the same time, the multimodal large models built for GUI Agents currently have poor understanding and decision-making abilities for Chinese GUI interfaces, making them difficult to apply to a large number of Chinese apps. This paper introduces MobileFlow, a multimodal large language model meticulously crafted for mobile GUI agents. Transforming from the open-source model Qwen-VL-Chat into GUI domain, MobileFlow contains approximately 21 billion parameters and is equipped with novel hybrid visual encoders, making it possible for variable resolutions of image inputs and good support for multilingual GUI. By incorporating Mixture of Experts (MoE) expansions and pioneering alignment training strategies, MobileFlow has the capacity to fully interpret image data and comprehend user instructions for GUI interaction tasks. Finally, MobileFlow outperforms Qwen-VL-Max and GPT-4v in terms of task execution by GUI agents on both public and our proposed evaluation metrics, and has been successfully deployed in real-world business contexts, proving its effectiveness for practical applications.

Xubo Zhu, Haoyang Zhang, Fei He et al.

3D semantic occupancy prediction is crucial for autonomous driving perception, offering comprehensive geometric scene understanding and semantic recognition. However, existing methods struggle with geometric misalignment in view transformation due to the lack of pixel-level accurate depth estimation, and severe spatial class imbalance where semantic categories exhibit strong spatial anisotropy. To address these challenges, we propose Dr.Occ, a depth- and region-guided occupancy prediction framework. Specifically, we introduce a depth-guided 2D-to-3D View Transformer (D$^2$-VFormer) that effectively leverages high-quality dense depth cues from MoGe-2 to construct reliable geometric priors, thereby enabling precise geometric alignment of voxel features. Moreover, inspired by the Mixture-of-Experts (MoE) framework, we propose a region-guided Expert Transformer (R/R$^2$-EFormer) that adaptively allocates region-specific experts to focus on different spatial regions, effectively addressing spatial semantic variations. Thus, the two components make complementary contributions: depth guidance ensures geometric alignment, while region experts enhance semantic learning. Experiments on the Occ3D-nuScenes benchmark demonstrate that \textbf{Dr.Occ} improves the strong baseline BEVDet4D by 7.43\% mIoU and 3.09\% IoU under the full vision-only setting.

Yifan Zhang, Rui Wu, Sergiu M. Dascalu et al.

Transformers have achieved remarkable performance in multivariate time series(MTS) forecasting due to their capability to capture long-term dependencies. However, the canonical attention mechanism has two key limitations: (1) its quadratic time complexity limits the sequence length, and (2) it generates future values from the entire historical sequence. To address this, we propose a Dozer Attention mechanism consisting of three sparse components: (1) Local, each query exclusively attends to keys within a localized window of neighboring time steps. (2) Stride, enables each query to attend to keys at predefined intervals. (3) Vary, allows queries to selectively attend to keys from a subset of the historical sequence. Notably, the size of this subset dynamically expands as forecasting horizons extend. Those three components are designed to capture essential attributes of MTS data, including locality, seasonality, and global temporal dependencies. Additionally, we present the Dozerformer Framework, incorporating the Dozer Attention mechanism for the MTS forecasting task. We evaluated the proposed Dozerformer framework with recent state-of-the-art methods on nine benchmark datasets and confirmed its superior performance. The experimental results indicate that excluding a subset of historical time steps from the time series forecasting process does not compromise accuracy while significantly improving efficiency. Code is available at https://github.com/GRYGY1215/Dozerformer.

Rui Wu, Ruixiang Tang

Reinforcement learning for LLMs is vulnerable to reward hacking, where models exploit shortcuts to maximize reward without solving the intended task. We systematically study this phenomenon in coding tasks using an environment-manipulation setting, where models can rewrite evaluator code to trivially pass tests without solving the task, as a controlled testbed. Across both studied models, we identify a reproducible three-phase rebound pattern: models first attempt to rewrite the evaluator but fail, as their rewrites embed test cases their own solutions cannot pass. They then temporarily retreat to legitimate solving. When legitimate reward remains scarce, they rebound into successful hacking with qualitatively different strategies. Using representation engineering, we extract concept directions for shortcut, deception, and evaluation awareness from domain-general contrastive pairs and find that the shortcut direction tracks hacking behavior most closely, making it an effective representational proxy for detection. Motivated by this finding, we propose Advantage Modification, which integrates shortcut concept scores into GRPO advantage computation to penalize hacking rollouts before policy updates. Because the penalty is internalized into the training signal rather than applied only at inference time, Advantage Modification provides more robust suppression of hacking compared with generation-time activation steering.

Rui Wu, Hong Xie, Yongjun Li

Current continuous generative models (e.g., Diffusion Models, Flow Matching) implicitly assume that locally consistent causal mechanisms naturally yield globally coherent counterfactuals. In this paper, we prove that this assumption fails fundamentally when the causal graph exhibits non-trivial homology (e.g., structural conflicts or hidden confounders). We formalize structural causal models as cellular sheaves over Wasserstein spaces, providing a strict algebraic topological definition of cohomological obstructions in measure spaces. To ensure computational tractability and avoid deterministic singularities (which we define as manifold tearing), we introduce entropic regularization and derive the Entropic Wasserstein Causal Sheaf Laplacian, a novel system of coupled non-linear Fokker-Planck equations. Crucially, we prove an entropic pullback lemma for the first variation of pushforward measures. By integrating this with the Implicit Function Theorem (IFT) on Sinkhorn optimality conditions, we establish a direct algorithmic bridge to automatic differentiation (VJP), achieving O(1)-memory reverse-mode gradients strictly independent of the iteration horizon. Empirically, our framework successfully leverages thermodynamic noise to navigate topological barriers ("entropic tunneling") in high-dimensional scRNA-seq counterfactuals. Finally, we invert this theoretical framework to introduce the Topological Causal Score, demonstrating that our Sheaf Laplacian acts as a highly sensitive algebraic detector for topology-aware causal discovery.

Zixun Lan, Maochun Xu, Yifan Ren et al.

Recent advances in large language models (LLMs) have led to substantial progress in domain-specific applications, particularly within the legal domain. However, general-purpose models such as GPT-4 often struggle with specialized subdomains that require precise legal knowledge, complex reasoning, and contextual sensitivity. To address these limitations, we present LabourLawLLM, a legal large language model tailored to Chinese labor law. We also introduce LabourLawBench, a comprehensive benchmark covering diverse labor-law tasks, including legal provision citation, knowledge-based question answering, case classification, compensation computation, named entity recognition, and legal case analysis. Our evaluation framework combines objective metrics (e.g., ROUGE-L, accuracy, F1, and soft-F1) with subjective assessment based on GPT-4 scoring. Experiments show that LabourLawLLM consistently outperforms general-purpose and existing legal-specific LLMs across task categories. Beyond labor law, our methodology provides a scalable approach for building specialized LLMs in other legal subfields, improving accuracy, reliability, and societal value of legal AI applications.

Rui Wu, Hong Xie

Continuous causal discovery typically couples representation learning with structural optimization via non-convex acyclicity penalties, which subjects solvers to local optima and restricts scalability in high-dimensional regimes. We propose a decoupled paradigm that shifts the causal discovery bottleneck from non-convex optimization to statistical score estimation. We introduce the Score-Schur Topological Sort (SSTS), an algorithm that extracts topological order directly from unconstrained generative models, bypassing constrained structure optimization. We establish that the causal hierarchy leaves a geometric signature within the score function: iterative graph marginalization is mathematically equivalent to computing the Schur complement of the Score-Jacobian Information Matrix (SJIM) under linear conditions. This translates the acyclicity constraint into an algebraic procedure with a dominant cost of O(d^3) operations. For non-linear systems, we formulate the expectation gap of Schur marginalization and introduce Block-SSTS to compress extraction depth, bounding structural error. Empirically, SSTS allows causal structural analysis on non-linear graphs up to d=1000. At this scale, our framework indicates that once the non-convex optimization bottleneck is mathematically bypassed, the structural fidelity of continuous causal discovery is bounded by the finite-sample estimation variance of the global score geometry. By reducing graph extraction to matrix operations, this work reframes scalable causal discovery from a constrained optimization problem to a statistical estimation challenge.

Rui Wu, Lizheng Wang, Yongjun Li

Judea Pearl's vision of Structural Causal Models (SCMs) as engines for counterfactual reasoning hinges on faithful abduction: the precise inference of latent exogenous noise. For decades, operationalizing this step for complex, non-linear mechanisms has remained a significant computational challenge. The advent of diffusion models, powerful universal function approximators, offers a promising solution. However, we argue that their standard design, optimized for perceptual generation over logical inference, introduces a fundamental flaw for this classical problem: an inherent information loss we term the Structural Reconstruction Error (SRE). To address this challenge, we formalize the principle of Causal Information Conservation (CIC) as the necessary condition for faithful abduction. We then introduce BELM-MDCM, the first diffusion-based framework engineered to be causally sound by eliminating SRE by construction through an analytically invertible mechanism. To operationalize this framework, a Targeted Modeling strategy provides structural regularization, while a Hybrid Training Objective instills a strong causal inductive bias. Rigorous experiments demonstrate that our Zero-SRE framework not only achieves state-of-the-art accuracy but, more importantly, enables the high-fidelity, individual-level counterfactuals required for deep causal inquiries. Our work provides a foundational blueprint that reconciles the power of modern generative models with the rigor of classical causal theory, establishing a new and more rigorous standard for this emerging field.

Can Jin, Rui Wu, Tong Che et al.

Ensuring that Large Language Models (LLMs) adhere to safety principles without refusing benign requests remains a significant challenge. While OpenAI introduces deliberative alignment (DA) to enhance the safety of its o-series models through reasoning over detailed ``code-like'' safety rules, the effectiveness of this approach in open-source LLMs, which typically lack advanced reasoning capabilities, is understudied. In this work, we systematically evaluate the impact of explicitly specifying extensive safety codes versus demonstrating them through illustrative cases. We find that referencing explicit codes inconsistently improves harmlessness and systematically degrades helpfulness, whereas training on case-augmented simple codes yields more robust and generalized safety behaviors. By guiding LLMs with case-augmented reasoning instead of extensive code-like safety rules, we avoid rigid adherence to narrowly enumerated rules and enable broader adaptability. Building on these insights, we propose CADA, a case-augmented deliberative alignment method for LLMs utilizing reinforcement learning on self-generated safety reasoning chains. CADA effectively enhances harmlessness, improves robustness against attacks, and reduces over-refusal while preserving utility across diverse benchmarks, offering a practical alternative to rule-only DA for improving safety while maintaining helpfulness.

Changli Wang, Rui Wu, Fang Yin

Human emotions are complex, with sarcasm being a subtle and distinctive form. Despite progress in sarcasm research, sarcasm generation remains underexplored, primarily due to the overreliance on textual modalities and the neglect of visual cues, as well as the mismatch between image content and sarcastic intent in existing datasets. In this paper, we introduce M2SaG, a multimodal sarcasm generation dataset with 4,970 samples, each containing an image, a sarcastic text, and a sarcasm target. To benchmark M2SaG, we propose ViSP, a generation framework that integrates Proximal Policy Optimization (PPO) and contrastive learning. PPO utilizes reward scores from DIP to steer the generation of sarcastic texts, while contrastive learning encourages the model to favor outputs with higher reward scores. These strategies improve overall generation quality and produce texts with more pronounced sarcastic intent. We evaluate ViSP across five metric sets and find it surpasses all baselines, including large language models, underscoring their limitations in sarcasm generation. Furthermore, we analyze the distributions of Sarcasm Scores and Factual Incongruity for both M2SaG and the texts generated by ViSP. The generated texts exhibit higher mean Sarcasm Scores (0.898 vs. 0.770) and Factual Incongruity (0.768 vs. 0.739), demonstrating that ViSP produces higher-quality sarcastic content than the original dataset. % The dataset and code will be publicly available. Our dataset and code will be released at \textit{https://github.com/wclapply/ViSP}.

Shaobo Guo, Xiao Jiang, Zhizhong Su et al.

As a critical cue for understanding human intention, human gaze provides a key signal for Human-Computer Interaction(HCI) applications. Appearance-based gaze estimation, which directly regresses the gaze vector from eye images, has made great progress recently based on Convolutional Neural Networks(ConvNets) architecture and open-source large-scale gaze datasets. However, encoding model-based knowledge into CNN model to further improve the gaze estimation performance remains a topic that needs to be explored. In this paper, we propose HybridGazeNet(HGN), a unified framework that encodes the geometric eyeball model into the appearance-based CNN architecture explicitly. Composed of a multi-branch network and an uncertainty module, HybridGazeNet is trained using a hyridized strategy. Experiments on multiple challenging gaze datasets shows that HybridGazeNet has better accuracy and generalization ability compared with existing SOTA methods. The code will be released later.

Yuxin Fang, Bencheng Liao, Xinggang Wang et al.

Can Transformer perform 2D object- and region-level recognition from a pure sequence-to-sequence perspective with minimal knowledge about the 2D spatial structure? To answer this question, we present You Only Look at One Sequence (YOLOS), a series of object detection models based on the vanilla Vision Transformer with the fewest possible modifications, region priors, as well as inductive biases of the target task. We find that YOLOS pre-trained on the mid-sized ImageNet-1k dataset only can already achieve quite competitive performance on the challenging COCO object detection benchmark, e.g., YOLOS-Base directly adopted from BERT-Base architecture can obtain 42.0 box AP on COCO val. We also discuss the impacts as well as limitations of current pre-train schemes and model scaling strategies for Transformer in vision through YOLOS. Code and pre-trained models are available at https://github.com/hustvl/YOLOS.

Yanhao Wu, Haoyang Zhang, Fei He et al.

End-to-end autonomous driving has rapidly progressed, enabling joint perception and planning in complex environments. In the planning stage, state-of-the-art (SOTA) end-to-end autonomous driving models decouple planning into parallel lateral and longitudinal predictions. While effective, this parallel design can lead to i) coordination failures between the planned path and speed, and ii) underutilization of the drive path as a prior for longitudinal planning, thus redundantly encoding static information. To address this, we propose a novel cascaded framework that explicitly conditions longitudinal planning on the drive path, enabling coordinated and collision-aware lateral and longitudinal planning. Specifically, we introduce a path-conditioned formulation that explicitly incorporates the drive path into longitudinal planning. Building on this, the model predicts longitudinal displacements along the drive path rather than full 2D trajectory waypoints. This design simplifies longitudinal reasoning and more tightly couples it with lateral planning. Additionally, we introduce a planning-oriented data augmentation strategy that simulates rare safety-critical events, such as vehicle cut-ins, by adding agents and relabeling longitudinal targets to avoid collision. Evaluated on the challenging Bench2Drive benchmark, our method sets a new SOTA, achieving a driving score of 89.07 and a success rate of 73.18%, demonstrating significantly improved coordination and safety

Yu Li, Shenyu Zhang, Rui Wu et al.

Recent advancements in generative Large Language Models(LLMs) have been remarkable, however, the quality of the text generated by these models often reveals persistent issues. Evaluating the quality of text generated by these models, especially in open-ended text, has consistently presented a significant challenge. Addressing this, recent work has explored the possibility of using LLMs as evaluators. While using a single LLM as an evaluation agent shows potential, it is filled with significant uncertainty and instability. To address these issues, we propose the MATEval: A "Multi-Agent Text Evaluation framework" where all agents are played by LLMs like GPT-4. The MATEval framework emulates human collaborative discussion methods, integrating multiple agents' interactions to evaluate open-ended text. Our framework incorporates self-reflection and Chain-of-Thought (CoT) strategies, along with feedback mechanisms, enhancing the depth and breadth of the evaluation process and guiding discussions towards consensus, while the framework generates comprehensive evaluation reports, including error localization, error types and scoring. Experimental results show that our framework outperforms existing open-ended text evaluation methods and achieves the highest correlation with human evaluation, which confirms the effectiveness and advancement of our framework in addressing the uncertainties and instabilities in evaluating LLMs-generated text. Furthermore, our framework significantly improves the efficiency of text evaluation and model iteration in industrial scenarios.

Huyue Ma, Yurui Jin, Helmut Hauser et al.

Due to brain-body co-evolution, animals' intrinsic body dynamics play a crucial role in energy-efficient locomotion, which shares control effort between active muscles and passive body dynamics -- a principle known as Embodied Physical Intelligence. In contrast, robot bodies are often designed with one centralised controller that typically suppress the intrinsic body dynamics instead of exploiting it. We introduce Physical Imitation Learning (PIL), which distils a Reinforcement Learning (RL) control policy into physically implementable body responses that can be directly offloaded to passive Parallel Elastic Joints (PEJs), enabling therefore the body to imitate part of the controlled behaviour. Meanwhile, the residual policy commands the motors to recover the RL policy's performance. The results is an overall reduced energy consumption thanks to outsourcing parts of the control policy to the PEJs. Here we show in simulated quadrupeds, that our PIL approach can offloads up to 87% of mechanical power to PEJs on flat terrain and 18% on rough terrain. Because the body design is distilled from -- rather than jointly optimised with -- the control policy, PIL realises brain-body co-design without expanding the search space with body design parameters, providing a computationally efficient route to task-specific Embodied Physical Intelligence applicable to a wide range of joint-based robot morphologies.

Rui Wu, Hong Xie, Yongjun Li

Judea Pearl's do-calculus provides a foundation for causal inference, but its translation to continuous generative models remains fraught with geometric challenges. We establish the fundamental limits of such interventions. We define the Counterfactual Event Horizon and prove the Manifold Tearing Theorem: deterministic flows inevitably develop finite-time singularities under extreme interventions. We establish the Causal Uncertainty Principle for the trade-off between intervention extremity and identity preservation. Finally, we introduce Geometry-Aware Causal Flow (GACF), a scalable algorithm that utilizes a topological radar to bypass manifold tearing, validated on high-dimensional scRNA-seq data.

Rui Wu, Yongjun Li

Continuous optimization has significantly advanced causal discovery, yet existing methods (e.g., NOTEARS) generally guarantee only asymptotic convergence to a stationary point. This often yields dense weighted matrices that require arbitrary post-hoc thresholding to recover a DAG. This gap between continuous optimization and discrete graph structures remains a fundamental challenge. In this paper, we bridge this gap by proposing the Hybrid-Order Acyclicity Constraint (AHOC) and optimizing it via the Smoothed Proximal Gradient (SPG-AHOC). Leveraging the Manifold Identification Property of proximal algorithms, we provide a rigorous theoretical guarantee: the Finite-Time Oracle Property. We prove that under standard identifiability assumptions, SPG-AHOC recovers the exact DAG support (structure) in finite iterations, even when optimizing a smoothed approximation. This result eliminates structural ambiguity, as our algorithm returns graphs with exact zero entries without heuristic truncation. Empirically, SPG-AHOC achieves state-of-the-art accuracy and strongly corroborates the finite-time identification theory.

Rui Wu, Li YongJun

Generative modeling typically seeks the path of least action via deterministic flows (ODE). While effective for in-distribution tasks, we argue that these deterministic paths become brittle under causal interventions, which often require transporting probability mass across low-density regions ("off-manifold") where the vector field is ill-defined. This leads to numerical instability and spurious correlations. In this work, we introduce the Causal Schrödinger Bridge (CSB), a framework that reformulates counterfactual inference as Entropic Optimal Transport. Unlike deterministic approaches that require strict invertibility, CSB leverages diffusion processes (SDEs) to robustly "tunnel" through support mismatches while strictly enforcing structural admissibility constraints. We prove the Structural Decomposition Theorem, showing that the global high-dimensional bridge factorizes into local, robust transitions. Empirical validation on high-dimensional interventions (Morpho-MNIST) demonstrates that CSB significantly outperforms deterministic baselines in structural consistency, particularly in regimes of strong, out-of-distribution treatments.

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.

Yanhao Wu, Haoyang Zhang, Tianwei Lin et al.

Generative models in Autonomous Driving (AD) enable diverse scene creation, yet existing methods fall short by only capturing a limited range of modalities, restricting the capability of generating controllable scenes for comprehensive evaluation of AD systems. In this paper, we introduce a multimodal generation framework that incorporates four major data modalities, including a novel addition of map modality. With tokenized modalities, our scene sequence generation framework autoregressively predicts each scene while managing computational demands through a two-stage approach. The Temporal AutoRegressive (TAR) component captures inter-frame dynamics for each modality while the Ordered AutoRegressive (OAR) component aligns modalities within each scene by sequentially predicting tokens in a fixed order. To maintain coherence between map and ego-action modalities, we introduce the Action-aware Map Alignment (AMA) module, which applies a transformation based on the ego-action to maintain coherence between these modalities. Our framework effectively generates complex, realistic driving scenes over extended sequences, ensuring multimodal consistency and offering fine-grained control over scene elements. Project page: https://yanhaowu.github.io/UMGen/

Rui Wu, Nikola Kovachki, Burigede Liu

Recent developments in mechanical, aerospace, and structural engineering have driven a growing need for efficient ways to model and analyse structures at much larger and more complex scales than before. While established numerical methods like the Finite Element Method remain reliable, they often struggle with computational cost and scalability when dealing with large and geometrically intricate problems. In recent years, neural network-based methods have shown promise because of their ability to efficiently approximate nonlinear mappings. However, most existing neural approaches are still largely limited to simple domains, which makes it difficult to apply to real-world PDEs involving complex geometries. In this paper, we propose a learning-based domain decomposition method (L-DDM) that addresses this gap. Our approach uses a single, pre-trained neural operator-originally trained on simple domains-as a surrogate model within a domain decomposition scheme, allowing us to tackle large and complicated domains efficiently. We provide a general theoretical result on the existence of neural operator approximations in the context of domain decomposition solution of abstract PDEs. We then demonstrate our method by accurately approximating solutions to elliptic PDEs with discontinuous microstructures in complex geometries, using a physics-pretrained neural operator (PPNO). Our results show that this approach not only outperforms current state-of-the-art methods on these challenging problems, but also offers resolution-invariance and strong generalization to microstructural patterns unseen during training.

Shenyu Zhang, Yu Li, Rui Wu et al.

Automatic methods for evaluating machine-generated texts hold significant importance due to the expanding applications of generative systems. Conventional methods tend to grapple with a lack of explainability, issuing a solitary numerical score to signify the assessment outcome. Recent advancements have sought to mitigate this limitation by incorporating large language models (LLMs) to offer more detailed error analyses, yet their applicability remains constrained, particularly in industrial contexts where comprehensive error coverage and swift detection are paramount. To alleviate these challenges, we introduce DEE, a Dual-stage Explainable Evaluation method for estimating the quality of text generation. Built upon Llama 2, DEE follows a dual-stage principle guided by stage-specific instructions to perform efficient identification of errors in generated texts in the initial stage and subsequently delves into providing comprehensive diagnostic reports in the second stage. DEE is fine-tuned on our elaborately assembled dataset AntEval, which encompasses 15K examples from 4 real-world applications of Alipay that employ generative systems. The dataset concerns newly emerged issues like hallucination and toxicity, thereby broadening the scope of DEE's evaluation criteria. Experimental results affirm that DEE's superiority over existing evaluation methods, achieving significant improvements in both human correlation as well as efficiency.

Rui Wu, Yihao Quan, Zeru Shi et al.

Safety-aligned Large Language Models (LLMs) still show two dominant failure modes: they are easily jailbroken, or they over-refuse harmless inputs that contain sensitive surface signals. We trace both to a common cause: current models reason weakly about links between actions and outcomes and over-rely on surface-form signals, lexical or stylistic cues that do not encode consequences. We define this failure mode as Consequence-blindness. To study consequence-blindness, we build a benchmark named CB-Bench covering four risk scenarios that vary whether semantic risk aligns with outcome risk, enabling evaluation under both matched and mismatched conditions which are often ignored by existing safety benchmarks. Mainstream models consistently fail to separate these risks and exhibit consequence-blindness, indicating that consequence-blindness is widespread and systematic. To mitigate consequence-blindness, we introduce CS-Chain-4k, a consequence-reasoning dataset for safety alignment. Models fine-tuned on CS-Chain-4k show clear gains against semantic-camouflage jailbreaks and reduce over-refusal on harmless inputs, while maintaining utility and generalization on other benchmarks. These results clarify the limits of current alignment, establish consequence-aware reasoning as a core alignment goal and provide a more practical and reproducible evaluation path.

Changli Wang, Fang Yin, Jiafeng Liu et al.

Visual and semantic concepts are often structured in a hierarchical manner. For instance, textual concept `cat' entails all images of cats. A recent study, MERU, successfully adapts multimodal learning techniques from Euclidean space to hyperbolic space, effectively capturing the visual-semantic hierarchy. However, a critical question remains: how can we more efficiently train a model to capture and leverage this hierarchy? In this paper, we propose the Hyperbolic Masked Image and Distillation Network (HMID-Net), a novel and efficient method that integrates Masked Image Modeling (MIM) and knowledge distillation techniques within hyperbolic space. To the best of our knowledge, this is the first approach to leverage MIM and knowledge distillation in hyperbolic space to train highly efficient models. In addition, we introduce a distillation loss function specifically designed to facilitate effective knowledge transfer in hyperbolic space. Our experiments demonstrate that MIM and knowledge distillation techniques in hyperbolic space can achieve the same remarkable success as in Euclidean space. Extensive evaluations show that our method excels across a wide range of downstream tasks, significantly outperforming existing models like MERU and CLIP in both image classification and retrieval.

Rui Wu, Zhaopeng Qiu, Jiacheng Jiang et al.

Medication recommendation targets to provide a proper set of medicines according to patients' diagnoses, which is a critical task in clinics. Currently, the recommendation is manually conducted by doctors. However, for complicated cases, like patients with multiple diseases at the same time, it's difficult to propose a considerate recommendation even for experienced doctors. This urges the emergence of automatic medication recommendation which can help treat the diagnosed diseases without causing harmful drug-drug interactions.Due to the clinical value, medication recommendation has attracted growing research interests.Existing works mainly formulate medication recommendation as a multi-label classification task to predict the set of medicines. In this paper, we propose the Conditional Generation Net (COGNet) which introduces a novel copy-or-predict mechanism to generate the set of medicines. Given a patient, the proposed model first retrieves his or her historical diagnoses and medication recommendations and mines their relationship with current diagnoses. Then in predicting each medicine, the proposed model decides whether to copy a medicine from previous recommendations or to predict a new one. This process is quite similar to the decision process of human doctors. We validate the proposed model on the public MIMIC data set, and the experimental results show that the proposed model can outperform state-of-the-art approaches.

Ding Li, Rui Wu, Yongqiang Tang et al.

Video moment retrieval aims to search the moment most relevant to a given language query. However, most existing methods in this community often require temporal boundary annotations which are expensive and time-consuming to label. Hence weakly supervised methods have been put forward recently by only using coarse video-level label. Despite effectiveness, these methods usually process moment candidates independently, while ignoring a critical issue that the natural temporal dependencies between candidates in different temporal scales. To cope with this issue, we propose a Multi-scale 2D Representation Learning method for weakly supervised video moment retrieval. Specifically, we first construct a two-dimensional map for each temporal scale to capture the temporal dependencies between candidates. Two dimensions in this map indicate the start and end time points of these candidates. Then, we select top-K candidates from each scale-varied map with a learnable convolutional neural network. With a newly designed Moments Evaluation Module, we obtain the alignment scores of the selected candidates. At last, the similarity between captions and language query is served as supervision for further training the candidates' selector. Experiments on two benchmark datasets Charades-STA and ActivityNet Captions demonstrate that our approach achieves superior performance to state-of-the-art results.

Yuxin Fang, Xinggang Wang, Rui Wu et al.

Recent studies indicate that hierarchical Vision Transformer with a macro architecture of interleaved non-overlapped window-based self-attention \& shifted-window operation is able to achieve state-of-the-art performance in various visual recognition tasks, and challenges the ubiquitous convolutional neural networks (CNNs) using densely slid kernels. Most follow-up works attempt to replace the shifted-window operation with other kinds of cross-window communication paradigms, while treating self-attention as the de-facto standard for window-based information aggregation. In this manuscript, we question whether self-attention is the only choice for hierarchical Vision Transformer to attain strong performance, and the effects of different kinds of cross-window communication. To this end, we replace self-attention layers with embarrassingly simple linear mapping layers, and the resulting proof-of-concept architecture termed as LinMapper can achieve very strong performance in ImageNet-1k image recognition. Moreover, we find that LinMapper is able to better leverage the pre-trained representations from image recognition and demonstrates excellent transfer learning properties on downstream dense prediction tasks such as object detection and instance segmentation. We also experiment with other alternatives to self-attention for content aggregation inside each non-overlapped window under different cross-window communication approaches, which all give similar competitive results. Our study reveals that the \textbf{macro architecture} of Swin model families, other than specific aggregation layers or specific means of cross-window communication, may be more responsible for its strong performance and is the real challenger to the ubiquitous CNN's dense sliding window paradigm. Code and models will be publicly available to facilitate future research.

Kun Zhang, Rui Wu, Ping Yao et al.

The target of 2D human pose estimation is to locate the keypoints of body parts from input 2D images. State-of-the-art methods for pose estimation usually construct pixel-wise heatmaps from keypoints as labels for learning convolution neural networks, which are usually initialized randomly or using classification models on ImageNet as their backbones. We note that 2D pose estimation task is highly dependent on the contextual relationship between image patches, thus we introduce a self-supervised method for pretraining 2D pose estimation networks. Specifically, we propose Heatmap-Style Jigsaw Puzzles (HSJP) problem as our pretext-task, whose target is to learn the location of each patch from an image composed of shuffled patches. During our pretraining process, we only use images of person instances in MS-COCO, rather than introducing extra and much larger ImageNet dataset. A heatmap-style label for patch location is designed and our learning process is in a non-contrastive way. The weights learned by HSJP pretext task are utilised as backbones of 2D human pose estimator, which are then finetuned on MS-COCO human keypoints dataset. With two popular and strong 2D human pose estimators, HRNet and SimpleBaseline, we evaluate mAP score on both MS-COCO validation and test-dev datasets. Our experiments show that downstream pose estimators with our self-supervised pretraining obtain much better performance than those trained from scratch, and are comparable to those using ImageNet classification models as their initial backbones.

Han Fu, Rui Wu, Chenghao Liu et al.

Nowadays, driven by the increasing concern on diet and health, food computing has attracted enormous attention from both industry and research community. One of the most popular research topics in this domain is Food Retrieval, due to its profound influence on health-oriented applications. In this paper, we focus on the task of cross-modal retrieval between food images and cooking recipes. We present Modality-Consistent Embedding Network (MCEN) that learns modality-invariant representations by projecting images and texts to the same embedding space. To capture the latent alignments between modalities, we incorporate stochastic latent variables to explicitly exploit the interactions between textual and visual features. Importantly, our method learns the cross-modal alignments during training but computes embeddings of different modalities independently at inference time for the sake of efficiency. Extensive experimental results clearly demonstrate that the proposed MCEN outperforms all existing approaches on the benchmark Recipe1M dataset and requires less computational cost.

Ge Liu, Rui Wu, Heng-Tze Cheng et al.

Deep Reinforcement Learning (RL) is proven powerful for decision making in simulated environments. However, training deep RL model is challenging in real world applications such as production-scale health-care or recommender systems because of the expensiveness of interaction and limitation of budget at deployment. One aspect of the data inefficiency comes from the expensive hyper-parameter tuning when optimizing deep neural networks. We propose Adaptive Behavior Policy Sharing (ABPS), a data-efficient training algorithm that allows sharing of experience collected by behavior policy that is adaptively selected from a pool of agents trained with an ensemble of hyper-parameters. We further extend ABPS to evolve hyper-parameters during training by hybridizing ABPS with an adapted version of Population Based Training (ABPS-PBT). We conduct experiments with multiple Atari games with up to 16 hyper-parameter/architecture setups. ABPS achieves superior overall performance, reduced variance on top 25% agents, and equivalent performance on the best agent compared to conventional hyper-parameter tuning with independent training, even though ABPS only requires the same number of environmental interactions as training a single agent. We also show that ABPS-PBT further improves the convergence speed and reduces the variance.

Kun Zhang, Peng He, Ping Yao et al.

Recently, multi-resolution networks (such as Hourglass, CPN, HRNet, etc.) have achieved significant performance on pose estimation by combining feature maps of various resolutions. In this paper, we propose a Resolution-wise Attention Module (RAM) and Gradual Pyramid Refinement (GPR), to learn enhanced resolution-wise feature maps for precise pose estimation. Specifically, RAM learns a group of weights to represent the different importance of feature maps across resolutions, and the GPR gradually merges every two feature maps from low to high resolutions to regress final human keypoint heatmaps. With the enhanced resolution-wise features learnt by CNN, we obtain more accurate human keypoint locations. The efficacies of our proposed methods are demonstrated on MS-COCO dataset, achieving state-of-the-art performance with average precision of 77.7 on COCO val2017 set and 77.0 on test-dev2017 set without using extra human keypoint training dataset.

Eugene Ie, Chih-wei Hsu, Martin Mladenov et al.

We propose RecSim, a configurable platform for authoring simulation environments for recommender systems (RSs) that naturally supports sequential interaction with users. RecSim allows the creation of new environments that reflect particular aspects of user behavior and item structure at a level of abstraction well-suited to pushing the limits of current reinforcement learning (RL) and RS techniques in sequential interactive recommendation problems. Environments can be easily configured that vary assumptions about: user preferences and item familiarity; user latent state and its dynamics; and choice models and other user response behavior. We outline how RecSim offers value to RL and RS researchers and practitioners, and how it can serve as a vehicle for academic-industrial collaboration.

Rui Zhang, Zheng Zhu, Peng Li et al.

Human pose estimation has witnessed a significant advance thanks to the development of deep learning. Recent human pose estimation approaches tend to directly predict the location heatmaps, which causes quantization errors and inevitably deteriorates the performance within the reduced network output. Aim at solving it, we revisit the heatmap-offset aggregation method and propose the Offset-guided Network (OGN) with an intuitive but effective fusion strategy for both two-stages pose estimation and Mask R-CNN. For two-stages pose estimation, a greedy box generation strategy is also proposed to keep more necessary candidates while performing person detection. For mask R-CNN, ratio-consistent is adopted to improve the generalization ability of the network. State-of-the-art results on COCO and PoseTrack dataset verify the effectiveness of our offset-guided pose estimation and tracking.

Eugene Ie, Vihan Jain, Jing Wang et al.

Most practical recommender systems focus on estimating immediate user engagement without considering the long-term effects of recommendations on user behavior. Reinforcement learning (RL) methods offer the potential to optimize recommendations for long-term user engagement. However, since users are often presented with slates of multiple items - which may have interacting effects on user choice - methods are required to deal with the combinatorics of the RL action space. In this work, we address the challenge of making slate-based recommendations to optimize long-term value using RL. Our contributions are three-fold. (i) We develop SLATEQ, a decomposition of value-based temporal-difference and Q-learning that renders RL tractable with slates. Under mild assumptions on user choice behavior, we show that the long-term value (LTV) of a slate can be decomposed into a tractable function of its component item-wise LTVs. (ii) We outline a methodology that leverages existing myopic learning-based recommenders to quickly develop a recommender that handles LTV. (iii) We demonstrate our methods in simulation, and validate the scalability of decomposed TD-learning using SLATEQ in live experiments on YouTube.

Rui Wu, Jiaming Xu, R. Srikant et al.

Given a set of pairwise comparisons, the classical ranking problem computes a single ranking that best represents the preferences of all users. In this paper, we study the problem of inferring individual preferences, arising in the context of making personalized recommendations. In particular, we assume that there are $n$ users of $r$ types; users of the same type provide similar pairwise comparisons for $m$ items according to the Bradley-Terry model. We propose an efficient algorithm that accurately estimates the individual preferences for almost all users, if there are $r \max \{m, n\}\log m \log^2 n$ pairwise comparisons per type, which is near optimal in sample complexity when $r$ only grows logarithmically with $m$ or $n$. Our algorithm has three steps: first, for each user, compute the \emph{net-win} vector which is a projection of its $\binom{m}{2}$-dimensional vector of pairwise comparisons onto an $m$-dimensional linear subspace; second, cluster the users based on the net-win vectors; third, estimate a single preference for each cluster separately. The net-win vectors are much less noisy than the high dimensional vectors of pairwise comparisons and clustering is more accurate after the projection as confirmed by numerical experiments. Moreover, we show that, when a cluster is only approximately correct, the maximum likelihood estimation for the Bradley-Terry model is still close to the true preference.

Kai Zhu, Rui Wu, Lei Ying et al.

In this paper, we consider a popular model for collaborative filtering in recommender systems where some users of a website rate some items, such as movies, and the goal is to recover the ratings of some or all of the unrated items of each user. In particular, we consider both the clustering model, where only users (or items) are clustered, and the co-clustering model, where both users and items are clustered, and further, we assume that some users rate many items (information-rich users) and some users rate only a few items (information-sparse users). When users (or items) are clustered, our algorithm can recover the rating matrix with $ω(MK \log M)$ noisy entries while $MK$ entries are necessary, where $K$ is the number of clusters and $M$ is the number of items. In the case of co-clustering, we prove that $K^2$ entries are necessary for recovering the rating matrix, and our algorithm achieves this lower bound within a logarithmic factor when $K$ is sufficiently large. We compare our algorithms with a well-known algorithms called alternating minimization (AM), and a similarity score-based algorithm known as the popularity-among-friends (PAF) algorithm by applying all three to the MovieLens and Netflix data sets. Our co-clustering algorithm and AM have similar overall error rates when recovering the rating matrix, both of which are lower than the error rate under PAF. But more importantly, the error rate of our co-clustering algorithm is significantly lower than AM and PAF in the scenarios of interest in recommender systems: when recommending a few items to each user or when recommending items to users who only rated a few items (these users are the majority of the total user population). The performance difference increases even more when noise is added to the datasets.

Jiaming Xu, Rui Wu, Kai Zhu et al.

In standard clustering problems, data points are represented by vectors, and by stacking them together, one forms a data matrix with row or column cluster structure. In this paper, we consider a class of binary matrices, arising in many applications, which exhibit both row and column cluster structure, and our goal is to exactly recover the underlying row and column clusters by observing only a small fraction of noisy entries. We first derive a lower bound on the minimum number of observations needed for exact cluster recovery. Then, we propose three algorithms with different running time and compare the number of observations needed by them for successful cluster recovery. Our analytical results show smooth time-data trade-offs: one can gradually reduce the computational complexity when increasingly more observations are available.

Rui Wu, R. Srikant, Jian Ni

We consider the structure learning problem for graphical models that we call loosely connected Markov random fields, in which the number of short paths between any pair of nodes is small, and present a new conditional independence test based algorithm for learning the underlying graph structure. The novel maximization step in our algorithm ensures that the true edges are detected correctly even when there are short cycles in the graph. The number of samples required by our algorithm is C*log p, where p is the size of the graph and the constant C depends on the parameters of the model. We show that several previously studied models are examples of loosely connected Markov random fields, and our algorithm achieves the same or lower computational complexity than the previously designed algorithms for individual cases. We also get new results for more general graphical models, in particular, our algorithm learns general Ising models on the Erdos-Renyi random graph G(p, c/p) correctly with running time O(np^5).