arXiv subjects

Kai He

Publications and source records attributed to Kai He.

At least 37 records · Page 2Linked to original sources

arXiv2026

S3-CoT: Self-Sampled Succinct Reasoning Enables Efficient Chain-of-Thought LLMs

Large language models (LLMs) equipped with chain-of-thought (CoT) achieve strong performance and offer a window into LLM behavior. However, recent evidence suggests that improvements in CoT capabilities often come with redundant reasoning processes, motivating a key question: Can LLMs acquire a fast-thinking mode analogous to human System 1 reasoning? To explore this, our study presents a self-sampling framework based on activation steering for efficient CoT learning. Our method can induce style-aligned and variable-length reasoning traces from target LLMs themselves without any teacher guidance, thereby alleviating a central bottleneck of SFT-based methods-the scarcity of high-quality supervision data. Using filtered data by gold answers, we perform SFT for efficient CoT learning with (i) a human-like dual-cognitive system, and (ii) a progressive compression curriculum. Furthermore, we explore a self-evolution regime in which SFT is driven solely by prediction-consistent data of variable-length variants, eliminating the need for gold answers. Extensive experiments on math benchmarks, together with cross-domain generalization tests in medicine, show that our method yields stable improvements for both general and R1-style LLMs. Our data and model checkpoints can be found at https://github.com/DYR1/S3-CoT.

cs.CL↗

arXiv2026

A Survey of Quantized Graph Representation Learning: Connecting Graph Structures with Large Language Models

Recent years have witnessed rapid advances in graph representation learning, with the continuous embedding approach emerging as the dominant paradigm. However, such methods encounter issues regarding parameter efficiency, interpretability, and robustness. Thus, Quantized Graph Representation (QGR) learning has recently gained increasing interest, which represents the graph structure with discrete codes instead of conventional continuous embeddings. Given its analogous representation form to natural language, QGR also possesses the capability to seamlessly integrate graph structures with large language models (LLMs). As this emerging paradigm is still in its infancy yet holds significant promise, we undertake this thorough survey to promote its rapid future prosperity. We first present the background of the general quantization methods and their merits. Moreover, we provide an in-depth demonstration of current QGR studies from the perspectives of quantized strategies, training objectives, distinctive designs, knowledge graph quantization, and applications. We further explore the strategies for code dependence learning and integration with LLMs. At last, we give discussions and conclude future directions, aiming to provide a comprehensive picture of QGR and inspire future research.

cs.LG↗

arXiv2026

$A^3$-Bench: Benchmarking Memory-Driven Scientific Reasoning via Anchor and Attractor Activation

Scientific reasoning relies not only on logical inference but also on activating prior knowledge and experiential structures. Memory can efficiently reuse knowledge and enhance reasoning consistency and stability. However, existing benchmarks mainly evaluate final answers or step-by-step coherence, overlooking the \textit{memory-driven} mechanisms that underlie human reasoning, which involves activating anchors and attractors, then integrating them into multi-step inference. To address this gap, we propose $A^3$-Bench~ https://a3-bench.github.io, a benchmark designed to evaluate scientific reasoning through dual-scale memory-driven activation, grounded in Anchor and Attractor Activation. First, we annotate 2,198 science reasoning problems across domains using the SAPM process(subject, anchor & attractor, problem, and memory developing). Second, we introduce a dual-scale memory evaluation framework utilizing anchors and attractors, along with the AAUI(Anchor--Attractor Utilization Index) metric to measure memory activation rates. Finally, through experiments with various base models and paradigms, we validate $A^3$-Bench and analyze how memory activation impacts reasoning performance, providing insights into memory-driven scientific reasoning.

cs.AI↗

arXiv2025

MARS: Multi-Agent Adaptive Reasoning with Socratic Guidance for Automated Prompt Optimization

Large language models (LLMs) typically operate in a question-answering paradigm, where the quality of the input prompt critically affects the response. Automated Prompt Optimization (APO) aims to overcome the cognitive biases of manually crafted prompts and explore a broader prompt design space. However, existing APO methods often suffer from rigid template structures and inefficient exploration in the prompt space. To this end, we propose a Multi-Agent Adaptive Reasoning with Socratic guidance framework (MARS) for APO. MARS consists of five complementary agents and formulates the optimization process as a Partially Observable Markov Decision Process (POMDP), enabling adaptive prompt refinement through explicit state modeling and interactive feedback. Specifically, a Planner agent generates flexible optimization trajectories, a Teacher-Critic-Student triad engages in Socratic-style dialogue to iteratively optimize the prompt based on pseudo-gradient signals in the text space, and a Target agent executes the prompt in downstream tasks to provide performance feedback. MARS integrates reasoning, feedback, and state transition into a unified hidden-state evolution process, improving both the effectiveness and interpretability of optimization. Extensive experiments on multiple datasets demonstrate that MARS outperforms existing APO methods in terms of optimization performance, search efficiency, and interpretability.

cs.CL↗

arXiv2025

Robust Multimodal Sentiment Analysis via Double Information Bottleneck

Multimodal sentiment analysis has received significant attention across diverse research domains. Despite advancements in algorithm design, existing approaches suffer from two critical limitations: insufficient learning of noise-contaminated unimodal data, leading to corrupted cross-modal interactions, and inadequate fusion of multimodal representations, resulting in discarding discriminative unimodal information while retaining multimodal redundant information. To address these challenges, this paper proposes a Double Information Bottleneck (DIB) strategy to obtain a powerful, unified compact multimodal representation. Implemented within the framework of low-rank Renyi's entropy functional, DIB offers enhanced robustness against diverse noise sources and computational tractability for high-dimensional data, as compared to the conventional Shannon entropy-based methods. The DIB comprises two key modules: 1) learning a sufficient and compressed representation of individual unimodal data by maximizing the task-relevant information and discarding the superfluous information, and 2) ensuring the discriminative ability of multimodal representation through a novel attention bottleneck fusion mechanism. Consequently, DIB yields a multimodal representation that effectively filters out noisy information from unimodal data while capturing inter-modal complementarity. Extensive experiments on CMU-MOSI, CMU-MOSEI, CH-SIMS, and MVSA-Single validate the effectiveness of our method. The model achieves 47.4% accuracy under the Acc-7 metric on CMU-MOSI and 81.63% F1-score on CH-SIMS, outperforming the second-best baseline by 1.19%. Under noise, it shows only 0.36% and 0.29% performance degradation on CMU-MOSI and CMU-MOSEI respectively.

cs.AI↗

arXiv2025

Multi-Atlas Brain Network Classification through Consistency Distillation and Complementary Information Fusion

In the realm of neuroscience, identifying distinctive patterns associated with neurological disorders via brain networks is crucial. Resting-state functional magnetic resonance imaging (fMRI) serves as a primary tool for mapping these networks by correlating blood-oxygen-level-dependent (BOLD) signals across different brain regions, defined as regions of interest (ROIs). Constructing these brain networks involves using atlases to parcellate the brain into ROIs based on various hypotheses of brain division. However, there is no standard atlas for brain network classification, leading to limitations in detecting abnormalities in disorders. Some recent methods have proposed utilizing multiple atlases, but they neglect consistency across atlases and lack ROI-level information exchange. To tackle these limitations, we propose an Atlas-Integrated Distillation and Fusion network (AIDFusion) to improve brain network classification using fMRI data. AIDFusion addresses the challenge of utilizing multiple atlases by employing a disentangle Transformer to filter out inconsistent atlas-specific information and distill distinguishable connections across atlases. It also incorporates subject- and population-level consistency constraints to enhance cross-atlas consistency. Additionally, AIDFusion employs an inter-atlas message-passing mechanism to fuse complementary information across brain regions. Experimental results on four datasets of different diseases demonstrate the effectiveness and efficiency of AIDFusion compared to state-of-the-art methods. A case study illustrates AIDFusion extract patterns that are both interpretable and consistent with established neuroscience findings.

eess.IV↗

arXiv2025

GEM: Empowering MLLM for Grounded ECG Understanding with Time Series and Images

While recent multimodal large language models (MLLMs) have advanced automated ECG interpretation, they still face two key limitations: (1) insufficient multimodal synergy between time series signals and visual ECG representations, and (2) limited explainability in linking diagnoses to granular waveform evidence. We introduce GEM, the first MLLM unifying ECG time series, 12-lead ECG images and text for grounded and clinician-aligned ECG interpretation. GEM enables feature-grounded analysis, evidence-driven reasoning, and a clinician-like diagnostic process through three core innovations: a dual-encoder framework extracting complementary time series and image features, cross-modal alignment for effective multimodal understanding, and knowledge-guided instruction generation for generating high-granularity grounding data (ECG-Grounding) linking diagnoses to measurable parameters ($e.g.$, QRS/PR Intervals). Additionally, we propose the Grounded ECG Understanding task, a clinically motivated benchmark designed to comprehensively assess the MLLM's capability in grounded ECG understanding. Experimental results on both existing and our proposed benchmarks show GEM significantly improves predictive performance (CSN $7.4\% \uparrow$), explainability ($22.7\% \uparrow$), and grounding ($24.8\% \uparrow$), making it more suitable for real-world clinical applications. GitHub repository: https://github.com/lanxiang1017/GEM.git

cs.CL↗

arXiv2025

ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulation

Recent advances in large generative models have greatly enhanced both image editing and in-context image generation, yet a critical gap remains in ensuring physical consistency, where edited objects must remain coherent. This capability is especially vital for world simulation related tasks. In this paper, we present ChronoEdit, a framework that reframes image editing as a video generation problem. First, ChronoEdit treats the input and edited images as the first and last frames of a video, allowing it to leverage large pretrained video generative models that capture not only object appearance but also the implicit physics of motion and interaction through learned temporal consistency. Second, ChronoEdit introduces a temporal reasoning stage that explicitly performs editing at inference time. Under this setting, target frame is jointly denoised with reasoning tokens to imagine a plausible editing trajectory that constrains the solution space to physically viable transformations. The reasoning tokens are then dropped after a few steps to avoid the high computational cost of rendering a full video. To validate ChronoEdit, we introduce PBench-Edit, a new benchmark of image-prompt pairs for contexts that require physical consistency, and demonstrate that ChronoEdit surpasses state-of-the-art baselines in both visual fidelity and physical plausibility. Project page for code and models: https://research.nvidia.com/labs/toronto-ai/chronoedit

cs.CV↗

arXiv2025

Kernel Representation and Similarity Measure for Incomplete Data

Measuring similarity between incomplete data is a fundamental challenge in web mining, recommendation systems, and user behavior analysis. Traditional approaches either discard incomplete data or perform imputation as a preprocessing step, leading to information loss and biased similarity estimates. This paper presents the proximity kernel, a new similarity measure that directly computes similarity between incomplete data in kernel feature space without explicit imputation in the original space. The proposed method introduces data-dependent binning combined with proximity assignment to project data into a high-dimensional sparse representation that adapts to local density variations. For missing value handling, we propose a cascading fallback strategy to estimate missing feature distributions. We conduct clustering tasks on the proposed kernel representation across 12 real world incomplete datasets, demonstrating superior performance compared to existing methods while maintaining linear time complexity. All the code are available at https://anonymous.4open.science/r/proximity-kernel-2289.

cs.LG↗

arXiv2025

Anchoring Refusal Direction: Mitigating Safety Risks in Tuning via Projection Constraint

Instruction Fine-Tuning (IFT) has been widely adopted as an effective post-training strategy to enhance various abilities of Large Language Models (LLMs). However, prior studies have shown that IFT can significantly compromise LLMs' safety, particularly their ability to refuse malicious instructions, raising significant concerns. Recent research into the internal mechanisms of LLMs has identified the refusal direction (r-direction) in the hidden states, which plays a pivotal role in governing refusal behavior. Building on this insight, our study reveals that the r-direction tends to drift during training, which we identify as one of the causes of the associated safety risks. To mitigate such drift, our proposed ProCon method introduces a projection-constrained loss term that regularizes the projection magnitude of each training sample's hidden state onto the r-direction. Our initial analysis shows that applying an appropriate constraint can effectively mitigate the refusal direction drift and associated safety risks, but remains limited by overall performance barriers. To overcome this barrier, informed by our observation of early-stage sharp drift and a data-driven perspective, we introduce a warm-up strategy that emphasizes early-stage strong constraints and broaden the data distribution to strengthen constraint signals, leading to an enhanced ProCon method. Experimental results under various datasets, scenarios, and LLMs demonstrate that our method can significantly mitigate safety risks posed by IFT while preserving task performance gains. Even compared with strong baselines, our method consistently delivers superior overall performance. Crucially, our analysis indicates that ProCon can contribute to stabilizing the r-direction during training, while such an interpretability-driven exploration of LLMs' internal mechanisms lays a solid foundation for future safety research.

cs.CL↗

arXiv2025

LuxDiT: Lighting Estimation with Video Diffusion Transformer

Estimating scene lighting from a single image or video remains a longstanding challenge in computer vision and graphics. Learning-based approaches are constrained by the scarcity of ground-truth HDR environment maps, which are expensive to capture and limited in diversity. While recent generative models offer strong priors for image synthesis, lighting estimation remains difficult due to its reliance on indirect visual cues, the need to infer global (non-local) context, and the recovery of high-dynamic-range outputs. We propose LuxDiT, a novel data-driven approach that fine-tunes a video diffusion transformer to generate HDR environment maps conditioned on visual input. Trained on a large synthetic dataset with diverse lighting conditions, our model learns to infer illumination from indirect visual cues and generalizes effectively to real-world scenes. To improve semantic alignment between the input and the predicted environment map, we introduce a low-rank adaptation finetuning strategy using a collected dataset of HDR panoramas. Our method produces accurate lighting predictions with realistic angular high-frequency details, outperforming existing state-of-the-art techniques in both quantitative and qualitative evaluations.

cs.GR↗

arXiv2025

Beyond Prediction: Reinforcement Learning as the Defining Leap in Healthcare AI

Reinforcement learning (RL) marks a fundamental shift in how artificial intelligence is applied in healthcare. Instead of merely predicting outcomes, RL actively decides interventions with long term goals. Unlike traditional models that operate on fixed associations, RL systems learn through trial, feedback, and long-term reward optimization, introducing transformative possibilities and new risks. From an information fusion lens, healthcare RL typically integrates multi-source signals such as vitals, labs clinical notes, imaging and device telemetry using temporal and decision-level mechanisms. These systems can operate within centralized, federated, or edge architectures to meet real-time clinical constraints, and naturally span data, features and decision fusion levels. This survey explore RL's rise in healthcare as more than a set of tools, rather a shift toward agentive intelligence in clinical environments. We first structure the landscape of RL techniques including model-based and model-free methods, offline and batch-constrained approaches, and emerging strategies for reward specification and uncertainty calibration through the lens of healthcare constraints. We then comprehensively analyze RL applications spanning critical care, chronic disease, mental health, diagnostics, and robotic assistance, identifying their trends, gaps, and translational bottlenecks. In contrast to prior reviews, we critically analyze RL's ethical, deployment, and reward design challenges, and synthesize lessons for safe, human-aligned policy learning. This paper serves as both a a technical roadmap and a critical reflection of RL's emerging transformative role in healthcare AI not as prediction machinery, but as agentive clinical intelligence.

cs.LG↗

arXiv2025

AtomDiffuser: Time-Aware Degradation Modeling for Drift and Beam Damage in STEM Imaging

Scanning transmission electron microscopy (STEM) plays a critical role in modern materials science, enabling direct imaging of atomic structures and their evolution under external interferences. However, interpreting time-resolved STEM data remains challenging due to two entangled degradation effects: spatial drift caused by mechanical and thermal instabilities, and beam-induced signal loss resulting from radiation damage. These factors distort both geometry and intensity in complex, temporally correlated ways, making it difficult for existing methods to explicitly separate their effects or model material dynamics at atomic resolution. In this work, we present AtomDiffuser, a time-aware degradation modeling framework that disentangles sample drift and radiometric attenuation by predicting an affine transformation and a spatially varying decay map between any two STEM frames. Unlike traditional denoising or registration pipelines, our method leverages degradation as a physically heuristic, temporally conditioned process, enabling interpretable structural evolutions across time. Trained on synthetic degradation processes, AtomDiffuser also generalizes well to real-world cryo-STEM data. It further supports high-resolution degradation inference and drift alignment, offering tools for visualizing and quantifying degradation patterns that correlate with radiation-induced atomic instabilities.

cs.CV↗

arXiv2025

Information Bottleneck-based Causal Attention for Multi-label Medical Image Recognition

Multi-label classification (MLC) of medical images aims to identify multiple diseases and holds significant clinical potential. A critical step is to learn class-specific features for accurate diagnosis and improved interpretability effectively. However, current works focus primarily on causal attention to learn class-specific features, yet they struggle to interpret the true cause due to the inadvertent attention to class-irrelevant features. To address this challenge, we propose a new structural causal model (SCM) that treats class-specific attention as a mixture of causal, spurious, and noisy factors, and a novel Information Bottleneck-based Causal Attention (IBCA) that is capable of learning the discriminative class-specific attention for MLC of medical images. Specifically, we propose learning Gaussian mixture multi-label spatial attention to filter out class-irrelevant information and capture each class-specific attention pattern. Then a contrastive enhancement-based causal intervention is proposed to gradually mitigate the spurious attention and reduce noise information by aligning multi-head attention with the Gaussian mixture multi-label spatial. Quantitative and ablation results on Endo and MuReD show that IBCA outperforms all methods. Compared to the second-best results for each metric, IBCA achieves improvements of 6.35\% in CR, 7.72\% in OR, and 5.02\% in mAP for MuReD, 1.47\% in CR, and 1.65\% in CF1, and 1.42\% in mAP for Endo.

cs.CV↗

arXiv2025

Towards Unified Neurosymbolic Reasoning on Knowledge Graphs

Knowledge Graph (KG) reasoning has received significant attention in the fields of artificial intelligence and knowledge engineering, owing to its ability to autonomously deduce new knowledge and consequently enhance the availability and precision of downstream applications. However, current methods predominantly concentrate on a single form of neural or symbolic reasoning, failing to effectively integrate the inherent strengths of both approaches. Furthermore, the current prevalent methods primarily focus on addressing a single reasoning scenario, presenting limitations in meeting the diverse demands of real-world reasoning tasks. Unifying the neural and symbolic methods, as well as diverse reasoning scenarios in one model is challenging as there is a natural representation gap between symbolic rules and neural networks, and diverse scenarios exhibit distinct knowledge structures and specific reasoning objectives. To address these issues, we propose a unified neurosymbolic reasoning framework, namely Tunsr, for KG reasoning. Tunsr first introduces a consistent structure of reasoning graph that starts from the query entity and constantly expands subsequent nodes by iteratively searching posterior neighbors. Based on it, a forward logic message-passing mechanism is proposed to update both the propositional representations and attentions, as well as first-order logic (FOL) representations and attentions of each node. In this way, Tunsr conducts the transformation of merging multiple rules by merging possible relations at each step. Finally, the FARI algorithm is proposed to induce FOL rules by constantly performing attention calculations over the reasoning graph. Extensive experimental results on 19 datasets of four reasoning scenarios (transductive, inductive, interpolation, and extrapolation) demonstrate the effectiveness of Tunsr.

cs.AI↗

arXiv2025

UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting

We address the challenge of relighting a single image or video, a task that demands precise scene intrinsic understanding and high-quality light transport synthesis. Existing end-to-end relighting models are often limited by the scarcity of paired multi-illumination data, restricting their ability to generalize across diverse scenes. Conversely, two-stage pipelines that combine inverse and forward rendering can mitigate data requirements but are susceptible to error accumulation and often fail to produce realistic outputs under complex lighting conditions or with sophisticated materials. In this work, we introduce a general-purpose approach that jointly estimates albedo and synthesizes relit outputs in a single pass, harnessing the generative capabilities of video diffusion models. This joint formulation enhances implicit scene comprehension and facilitates the creation of realistic lighting effects and intricate material interactions, such as shadows, reflections, and transparency. Trained on synthetic multi-illumination data and extensive automatically labeled real-world videos, our model demonstrates strong generalization across diverse domains and surpasses previous methods in both visual fidelity and temporal consistency.

cs.CV↗

arXiv2025

DivScore: Zero-Shot Detection of LLM-Generated Text in Specialized Domains

Detecting LLM-generated text in specialized and high-stakes domains like medicine and law is crucial for combating misinformation and ensuring authenticity. However, current zero-shot detectors, while effective on general text, often fail when applied to specialized content due to domain shift. We provide a theoretical analysis showing this failure is fundamentally linked to the KL divergence between human, detector, and source text distributions. To address this, we propose DivScore, a zero-shot detection framework using normalized entropy-based scoring and domain knowledge distillation to robustly identify LLM-generated text in specialized domains. We also release a domain-specific benchmark for LLM-generated text detection in the medical and legal domains. Experiments on our benchmark show that DivScore consistently outperforms state-of-the-art detectors, with 14.4% higher AUROC and 64.0% higher recall (0.1% false positive rate threshold). In adversarial settings, DivScore demonstrates superior robustness than other baselines, achieving on average 22.8% advantage in AUROC and 29.5% in recall. Code and data are publicly available.

cs.CL↗

arXiv2025

BrainPrompt: Multi-Level Brain Prompt Enhancement for Neurological Condition Identification

Neurological conditions, such as Alzheimer's Disease, are challenging to diagnose, particularly in the early stages where symptoms closely resemble healthy controls. Existing brain network analysis methods primarily focus on graph-based models that rely solely on imaging data, which may overlook important non-imaging factors and limit the model's predictive power and interpretability. In this paper, we present BrainPrompt, an innovative framework that enhances Graph Neural Networks (GNNs) by integrating Large Language Models (LLMs) with knowledge-driven prompts, enabling more effective capture of complex, non-imaging information and external knowledge for neurological disease identification. BrainPrompt integrates three types of knowledge-driven prompts: (1) ROI-level prompts to encode the identity and function of each brain region, (2) subject-level prompts that incorporate demographic information, and (3) disease-level prompts to capture the temporal progression of disease. By leveraging these multi-level prompts, BrainPrompt effectively harnesses knowledge-enhanced multi-modal information from LLMs, enhancing the model's capability to predict neurological disease stages and meanwhile offers more interpretable results. We evaluate BrainPrompt on two resting-state functional Magnetic Resonance Imaging (fMRI) datasets from neurological disorders, showing its superiority over state-of-the-art methods. Additionally, a biomarker study demonstrates the framework's ability to extract valuable and interpretable information aligned with domain knowledge in neuroscience. The code is available at https://github.com/AngusMonroe/BrainPrompt

q-bio.NC↗