Search arXiv⌕ Search

arXiv subjects

Shuo Han

Publications and source records attributed to Shuo Han.

At least 19 recordsLinked to original sources
arXiv2026

Jailbreak Connectivity: Towards Diverse, Transferable, and Universal Multimodal Large Language Model Jailbreak

Multimodal large language models (MLLMs) are vulnerable to visual jailbreak attacks that bypass safety safeguards. Existing methods typically optimize isolated adversarial examples, limiting attack diversity, transferability, and generalization. We propose Jailbreak Connectivity (JC), a framework based on the observation that effective adversarial inputs can be connected through low-loss regions in image space. JC optimizes continuous adversarial paths: path sampling yields diverse jailbreak images, lightweight surrogate classifiers guide cross-model transfer, and a distribution-level objective enables universal attacks across harmful queries. On MM-SafetyBench, JC achieves average attack success rates of 84.01%, 85.77%, and 97.95% on Qwen3-VL-8B-Instruct, InternVL3.5-8B, and Gemma-3-12B-IT, respectively, exceeding the strongest baselines by up to 68.36 percentage points. In black-box evaluations, JC improves attack success rates over FigStep by 24.69 and 27.34 percentage points on GPT-5.6 and Gemini 3.1 Pro, respectively. These findings reveal structured vulnerabilities in MLLM safety alignment and inform more comprehensive security evaluation.

cs.LG↗
arXiv2026

ImageRSI: Verification-Driven Recursive Self-Improvement for Agentic Image Generation

Complex text-to-image requests remain difficult to satisfy reliably. Verifying and repairing an image can fix an individual output, but a successful repair does not necessarily carry over to other requests. We present ImageRSI, a framework that keeps the image generator frozen and combines three components: five-dimensional verification and repair within each task, an experience library shared across tasks, and offline updating of the judge. Candidate experiences enter the library only after paired A/B validation, and the judge is updated with supervised fine-tuning and GRPO on accumulated image-judgment records and evaluated before reuse. We also introduce ImageReq-Bench, a benchmark of 961 requests built around hard cases in real-world image-generation requests, each paired with an evaluation checklist. ImageRSI achieves a Checklist Strict Accuracy (CSA) of 52.6% on MindBench (with Nano Banana Pro as the generator) and 65.66% on ImageReq-Bench, and we compare system outcomes across judge-training configurations.

cs.CV↗
arXiv2026

Active Inference via Incentive Design: Eliciting Divergent Behaviors under Partial Observability

We introduce Active Inference through Incentive Design, a framework in which a leader offers strategic incentive to elicit diverging behaviors from followers of different types, thereby improving intent inference under imperfect observations. We analyze two settings: Case 1: a perfectly observable leader with two homogeneous followers with different original/unmodified reward functions, under the assumption that followers' best responses induce Gibbs trajectory distributions in their respective Markov decision processes, we show that this inference problem can be reformulated as maximizing the difference in expected cumulative return, under a modified reward for one follower, between the best-response policies of the two followers; Case 2 considers a partially observable leader who aims to distinguish among heterogeneous followers differing in dynamics and/or rewards. The leader balances incentive cost against information gain, measured by the posterior entropy over follower types, consider the follower's best responses to the incentive design. We develop an efficient first-order gradient-based algorithm to optimize the information-theoretic inference objective. Experiments in stochastic gridworld environments demonstrate significant improvements in both inference accuracy and efficiency over unincentivized baselines.

eess.SY↗
arXiv2026

Baselines and Score Centering in Policy Gradient Estimation

Reinforcement learning has become a widely used approach to post-training large language models. In a typical implementation, learning is off-policy: The policy generating the responses differs from the policy being trained because of training-inference mismatch or policy staleness in asynchronous training. This discrepancy is commonly addressed by importance sampling, whose ratio is usually regularized in practice through truncation, masking, or clipping. Under regularized importance sampling, the importance-weighted policy gradient is no longer invariant to constant shifts in the reward. This invariance is recovered by using the covariance between the reward and the importance-weighted score as the estimand, which coincides with the policy gradient when the ratio is not regularized. This note develops this covariance perspective for both sequence-level and token-level importance sampling. The covariance perspective reveals the roles of the group-mean baseline and score centering in policy gradient estimation. Both perform centering in covariance estimation, but they differ in the mean being subtracted: The group-mean baseline uses a sample mean, whereas score centering uses a population mean. The two lead to distinct unbiased covariance estimators with different variances. Whether both forms of centering should be used simultaneously depends on whether importance sampling is applied at the sequence level or at the token level. With sequence-level importance sampling, score centering adds no benefit once the group-mean baseline is used; with token-level importance sampling, using the group-mean baseline without score centering generally introduces bias when the importance ratio is regularized.

cs.LG↗
arXiv2026

Credit Where It Matters: Dependency-Aware Policy Optimization for Terminal Agents

Terminal-using agents benefit from reinforcement learning (RL) in coding, debugging, and other multi-step terminal tasks. In these tasks, later commands often depend on information or intermediate results produced by earlier commands. However, existing trajectory-level and step-level credit assignment methods do not explicitly trace the read-write dependencies through which commands affect the final outcome. Consequently, training signals could still be assigned to irrelevant operations, weakening learning from relevant steps. In this paper, we propose Dependency-Aware Group Policy Optimization (DepGPO), which uses execution dependencies between commands to guide credit assignment for terminal agents. Specifically, we construct a command dependency graph from execution traces and trace backward from the resources inspected by the task verifier. We then assign credit to relevant writes and their supporting reads along these paths, and use it to redistribute trajectory advantages across steps. Extensive comparative experiments and ablation studies demonstrate that DepGPO improves task performance and training stability on complex terminal tasks.

cs.AI↗
arXiv2026

PhGPO: Pheromone-Guided Policy Optimization for Long-Horizon Tool Planning

Recent advancements in Large Language Model (LLM) agents have demonstrated strong capabilities in executing complex tasks through tool use. However, long-horizon multi-step tool planning is challenging, because the exploration space suffers from a combinatorial explosion. In this scenario, even when a correct tool-use path is found, it is usually considered an immediate reward for current training, which would not provide any reusable information for subsequent training. In this paper, we argue that historically successful trajectories contain reusable tool-transition patterns, which can be leveraged throughout the whole training process. Inspired by ant colony optimization where historically successful paths can be reflected by the pheromone, we propose Pheromone-Guided Policy Optimization (PhGPO), which learns a trajectory-based transition pattern (i.e., pheromone) from historical trajectories and then uses the learned pheromone to guide policy optimization. This learned pheromone provides explicit and reusable guidance that steers policy optimization toward historically successful tool transitions, thereby improving long-horizon tool planning. Comprehensive experimental results demonstrate the effectiveness of our proposed PhGPO.

cs.AI↗
arXiv2026

Incentive Design for Multi-Agent Systems: A Bilevel Optimization Framework for Coordinating Independent Agents and Convergence Analysis

Incentive design aims to guide the performance of a system towards a human's intention or preference. We study this problem in a multi-agent system with one leader and multiple followers. Each follower independently solves a mdp to maximize its own expected total return with the same state space and action space. However, the leader's objective depends on the collective best-response policies of all followers. To influence these policies of followers, the leader provides side payments as incentives to individual followers at a cost, aiming to align the collective behaviors of followers with its own goal while minimizing this cost of incentive. Such a leader-followers interaction is formulated as a bilevel optimization problem: the lower level consists of followers individually optimizing their MDPs given the side payments, and the upper level involves the leader optimizing its objective function given the followers' best responses. The main challenge to solve the incentive design is that the leader's objective is generally non-concave and the lower level optimization problems can have multiple local optima. To this end, we employ a constrained optimization reformation of this bi-level optimization problem and develop an algorithm that provably converges to a stationary point of the original problem, by leveraging several smoothness properties of value functions in MDPs. We validate our algorithm in a stochastic gridworld by examining its convergence, verifying that the constraints are satisfied, and evaluating the improvement in the leader's performance.

eess.SY↗
arXiv2026

RIBOSPAN: A Long-Context RNA Foundation Model for Versatile RNA Modeling

Full-length RNAs, particularly messenger RNAs, often exceed the context lengths used to pretrain existing RNA foundation models, limiting complete-transcript modeling at single-nucleotide resolution. We present RIBOSPAN, a 1.61-billion-parameter bidirectional RNA foundation model natively pretrained with context lengths up to 10,240 nt. RIBOSPAN combines dense bidirectional self-attention, single-nucleotide tokenization, and attention-isolated sequence packing to enable high-resolution modeling of complete long RNAs. Native 10K pretraining preserves strong reconstruction at 10,240 tokens and, in a controlled long-context benchmark, maintains strong contextual responsiveness and context-specific representation separation while keeping perturbation-induced changes highly localized. Inference-time YaRN scaling recovers much of the contextual organization lost by direct short-context extrapolation, but induces substantially greater distal representation diffusion. Frozen RNA-type evaluations show that RIBOSPAN learns state-of-the-art RNA representations, with a particularly clear advantage on long RNAs. Across downstream biological benchmarks, RIBOSPAN emerges as the strongest encoder-only RNA foundation model, achieving state-of-the-art performance in both full-transcript biological property prediction and zero-shot mutation-fitness modeling. Building on the same backbone, we develop a multidimensionally conditioned discrete-diffusion framework for full-length mRNA generation and redesign, including synonymous-codon diffusion for protein-preserving CDS optimization. Together, RIBOSPAN establishes a powerful long-context foundation for transferable RNA representation learning, biological prediction, and full-transcript mRNA design.

cs.LG↗
arXiv2026

SkillMOO: Multi-Objective Optimization of Agent Skills for Software Engineering

Agent skills are increasingly used to configure coding agents for software engineering (SE) tasks, yet current practice treats them as static, hand-crafted assets, or evolved on pass rate alone. This is insufficient: a skill can improve task success while substantially raising token cost, or introducing misleading guidance. We argue that SE agent skill bundles can be treated as multi-objective search objects and present SkillMOO, a framework that evolves skill bundles through LLM-proposed edits and NSGA-II Pareto selection on pass rate and inference cost. Evaluated across all 16 SkillsBench SE tasks, SkillMOO achieves the top pass rate rank on 11 of 12 non-zero-pass tasks while achieving cost reductions of up to 31.7% over static bundles, with pass rate gains up to 21 percentage points. Analysis of 38 skill edits shows that pruning and substitution dominate successful operations, offering actionable principles for skill bundle design. Thereby, the current practice of deploying skills without cost-aware validation leaves better skill configurations unexplored, motivating a new class of cost-aware, search-based skill engineering.

cs.SE↗
arXiv2026

Pretrained Event Classification Model for High Energy Physics Analysis

We introduce a foundation model for event classification in high-energy physics, built on a Graph Neural Network architecture and trained on 120 million simulated proton-proton collision events spanning 12 distinct physics processes. The model is pretrained to learn a general and robust representation of collision data using challenging multiclass and multilabel classification tasks. Its performance is evaluated across seven event classification tasks, which include new physics processes not encountered during pretraining as well as ATLAS Open Data to demonstrate generalizability across different simulation frameworks, from Delphes fast simulation to full ATLAS detector simulation. Fine-tuning the pretrained model significantly improves classification performance, particularly in scenarios with limited training data, demonstrating gains in both accuracy and computational efficiency. To investigate the underlying mechanisms behind these performance improvements, we employ a representational similarity evaluation framework based on Centered Kernel Alignment. This analysis reveals that encoder-stage representations of the fine-tuned model remain similar to those of the baseline, while intermediate graph processing layers diverge substantially, indicating that fine-tuning preserves general-purpose encoders while developing fundamentally different message-passing pathways to arrive at superior task performance.

hep-ph↗
arXiv2026

Ratio-Variance Regularized Policy Optimization

Standard on-policy reinforcement learning relies on heuristic clipping to enforce trust regions, but this mechanism imposes a severe cost by indiscriminately truncating high-return yet high-divergence updates. We demonstrate that explicitly constraining the policy ratio variance provides a principled local approximation to trust-region constraints, eliminating the need for binary hard clipping. By acting as a distributional ``soft brake'', this approach preserves critical gradient signals from novel discoveries while naturally down-weighting and enabling the reuse of stale, off-policy data. We introduce ${\bf R}^2{\bf VPO}$ (Ratio-Variance Regularized Policy Optimization), which implements this constraint via a primal-dual optimization framework. Extensive evaluations across $7$ LLM scales, spanning both fast and slow reasoning paradigms, and $10$ robotic control tasks demonstrate the generality of the proposed approach. R$^2$VPO achieves substantial performance gains on mathematical reasoning benchmarks, with particularly pronounced improvements on smaller models, while significantly improving sample efficiency. Furthermore, it consistently outperforms PPO baselines in continuous control domains, particularly in sparse-reward and dynamic environments. Together, these findings establish ratio-variance regularization as a principled foundation for stable and data-efficient policy optimization.

cs.LG↗
arXiv2026

D2 Actor Critic: Diffusion Actor Meets Distributional Critic

We introduce D2AC, a new model-free reinforcement learning (RL) algorithm designed to train expressive diffusion policies online effectively. At its core is a policy improvement objective that avoids the high variance of typical policy gradients and the complexity of backpropagation through time. This stable learning process is critically enabled by our second contribution: a robust distributional critic, which we design through a fusion of distributional RL and clipped double Q-learning. The resulting algorithm is highly effective, achieving state-of-the-art performance on a benchmark of eighteen hard RL tasks, including Humanoid, Dog, and Shadow Hand domains, spanning both dense-reward and goal-conditioned RL scenarios. Beyond standard benchmarks, we also evaluate a biologically motivated predator-prey task to examine the behavioral robustness and generalization capacity of our approach. Code: https://github.com/d2ac-actor-critic/d2ac-public

cs.LG↗
arXiv2026

Medical thinking with multiple images

Large language models perform well on many medical QA benchmarks, but real clinical reasoning often requires integrating evidence across multiple images rather than interpreting a single view. We introduce MedThinkVQA, an expert-annotated benchmark for thinking with multiple images, where models must interpret each image, combine cross-view evidence, and answer diagnostic questions with intermediate supervision and step-level evaluation. The dataset contains 8,067 cases, including 720 test cases, with an average of 6.62 images per case, substantially denser than prior work, whose expert-level benchmarks use at most 1.43 images per case. On the test set, the best closed-source models, Claude-4.6-Opus, Gemini-3-Pro, and GPT-5.2-xhigh, reach only 57.2%, 55.3%, and 54.9% accuracy, while GPT-5-mini and GPT-5-nano reach 39.7% and 30.8%. Strong open-source models lag behind, led by Qwen3.5-397B-A17B at 52.2% and Qwen3.5-27B at 50.6%. Further analysis identifies grounded multi-image reasoning as the main bottleneck: models often fail to extract, align, and compose evidence across views before higher-level inference can help. Providing expert single-image cues and cross-image summaries improves performance, whereas replacing them with self-generated intermediates reduces accuracy. Step-level analysis shows that over 70% of errors arise from image reading and cross-view integration. Scaling results further show that additional inference-time computation helps only when visual grounding is already reliable; when early evidence extraction is weak, longer reasoning yields limited or unstable gains and can amplify misread cues. These results suggest that the key challenge is not reasoning length alone, but reliable mechanisms for grounding, aligning, and composing distributed evidence across real-world multimodal clinical inputs.

cs.CV↗
arXiv2026

Planning Stealthy Backdoor Attacks in MDPs with Observation-Based Triggers

This paper investigates backdoor attack planning in stochastic control systems modeled as Markov Decision Processes (MDPs). A backdoor attack involves an adversary deploying a policy that performs well in the original MDP to pass testing, but behaves maliciously at runtime when combined with a trigger that perturbs system dynamics. We consider a sophisticated attacker capable of jointly optimizing the backdoor policy and its trigger using only a blackbox simulator. During execution, the attacker has access only to partial observations of the system state and is restricted to introduce small perturbations to the system's transition dynamics. We formulate the attack planning problem as a constrained Markov game with an augmented state space and two players: Player 0 learns a backdoor policy that maximizes attack rewards when the trigger is active. However, when the trigger is inactive, the backdoor policy behaves near-optimally in the original MDP; Player 1 designs a finite-memory, observation-based trigger to activate the attack. We propose a switching gradient-based optimization algorithm to jointly solve for the backdoor policy and trigger. Experiments on a case study demonstrate the effectiveness of our method in achieving stealthy and successful backdoor attacks, and how the attack performance varies under different parameters related to the stealthiness of the backdoor attack.

eess.SY↗
arXiv2026

Cross-Distribution Diffusion Priors-Driven Iterative Reconstruction for Sparse-View CT

Sparse-View CT (SVCT) reconstruction enhances temporal resolution and reduces radiation dose, yet its clinical use is hindered by artifacts due to view reduction and domain shifts from scanner, protocol, or anatomical variations, leading to performance degradation in out-of-distribution (OOD) scenarios. In this work, we propose a Cross-Distribution Diffusion Priors-Driven Iterative Reconstruction (CDPIR) framework to tackle the OOD problem in SVCT. CDPIR integrates cross-distribution diffusion priors, derived from a Scalable Interpolant Transformer (SiT), with model-based iterative reconstruction methods. Specifically, we train a SiT backbone, an extension of the Diffusion Transformer (DiT) architecture, to establish a unified stochastic interpolant framework, leveraging Classifier-Free Guidance (CFG) across multiple datasets. By randomly dropping the conditioning with a null embedding during training, the model learns both domain-specific and domain-invariant priors, enhancing generalizability. During sampling, the globally sensitive transformer-based diffusion model exploits the cross-distribution prior within the unified stochastic interpolant framework, enabling flexible and stable control over multi-distribution-to-noise interpolation paths and decoupled sampling strategies, thereby improving adaptation to OOD reconstruction. By alternating between data fidelity and sampling updates, our model achieves state-of-the-art performance with superior detail preservation in SVCT reconstructions. Extensive experiments demonstrate that CDPIR significantly outperforms existing approaches, particularly under OOD conditions, highlighting its robustness and potential clinical value in challenging imaging scenarios.

eess.IV↗
arXiv2026

RiTeK: A Dataset for Large Language Models Complex Reasoning over Textual Knowledge Graphs in Medicine

Answering complex real-world questions in the medical domain often requires accurate retrieval from medical Textual Knowledge Graphs (medical TKGs), as the relational path information from TKGs could enhance the inference ability of Large Language Models (LLMs). However, the main bottlenecks lie in the scarcity of existing medical TKGs, the limited expressiveness of their topological structures, and the lack of comprehensive evaluations of current retrievers for medical TKGs. To address these challenges, we first develop a Dataset1 for LLMs Complex Reasoning over medical Textual Knowledge Graphs (RiTeK), covering a broad range of topological structures. Specifically, we synthesize realistic user queries integrating diverse topological structures, relational information, and complex textual descriptions. We conduct a rigorous medical expert evaluation process to assess and validate the quality of our synthesized queries. RiTeK also serves as a comprehensive benchmark dataset for evaluating the capabilities of retrieval systems built upon LLMs. By assessing 11 representative retrievers on this benchmark, we observe that existing methods struggle to perform well, revealing notable limitations in current LLM-driven retrieval approaches. These findings highlight the pressing need for more effective retrieval systems tailored for semi-structured data in the medical domain.

cs.CL↗
arXiv2026

ChatCLIDS: Simulating Persuasive AI Dialogues to Promote Closed-Loop Insulin Adoption in Type 1 Diabetes Care

Real-world adoption of closed-loop insulin delivery systems (CLIDS) in type 1 diabetes remains low, driven not by technical failure, but by diverse behavioral, psychosocial, and social barriers. We introduce ChatCLIDS, the first benchmark to rigorously evaluate LLM-driven persuasive dialogue for health behavior change. Our framework features a library of expert-validated virtual patients, each with clinically grounded, heterogeneous profiles and realistic adoption barriers, and simulates multi-turn interactions with nurse agents equipped with a diverse set of evidence-based persuasive strategies. ChatCLIDS uniquely supports longitudinal counseling and adversarial social influence scenarios, enabling robust, multi-dimensional evaluation. Our findings reveal that while larger and more reflective LLMs adapt strategies over time, all models struggle to overcome resistance, especially under realistic social pressure. These results highlight critical limitations of current LLMs for behavior change, and offer a high-fidelity, scalable testbed for advancing trustworthy persuasive AI in healthcare and beyond.

cs.AI↗
arXiv2026

Clinical Metadata Guided Limited-Angle CT Image Reconstruction

Limited-angle computed tomography (LACT) offers improved temporal resolution and reduced radiation dose for cardiac imaging, but suffers from severe artifacts due to truncated projections. To address the ill-posedness of LACT reconstruction, we propose a two-stage diffusion framework guided by structured clinical metadata. In the first stage, a transformer-based diffusion model conditioned exclusively on metadata, including acquisition parameters, patient demographics, and diagnostic impressions, generates coarse anatomical priors from noise. The second stage further refines the images by integrating both the coarse prior and metadata to produce high-fidelity results. Physics-based data consistency is enforced at each sampling step in both stages using an Alternating Direction Method of Multipliers module, ensuring alignment with the measured projections. Extensive experiments on both synthetic and real cardiac CT datasets demonstrate that incorporating metadata significantly improves reconstruction fidelity, particularly under severe angular truncation. Compared to existing metadata-free baselines, our method achieves superior performance in SSIM, PSNR, nMI, and PCC. Ablation studies confirm that different types of metadata contribute complementary benefits, particularly diagnostic and demographic priors under limited-angle conditions. These findings highlight the dual role of clinical metadata in improving both reconstruction quality and efficiency, supporting their integration into future metadata-guided medical imaging frameworks.

cs.CV↗