15 papers
Taming CoT Obfuscation in VLMs: From Mechanistic Evidence to Activation-Level Enforcement
Xutao Mao, Jianing Zhu, Jinman Zhao +4
Reinforcement learning (RL) has substantially improved the reasoning capabilities of vision-language models (VLMs), yet it can induce chain-of-thought (CoT) obfuscation, an operati…
Act More, Decide Less: Skill-Guided Adaptive Action Chunking for Long-Horizon LLM Agents
Yanting Yang, Can Jin, Jinman Zhao +6
Large language model (LLM) agents for long-horizon interactive tasks typically follow a ReAct-style protocol, issuing one primitive action per LLM round. While this enables frequen…
Efficient and Trainable Language Model Test-Time Scaling via Local Branch Routing
Yutong Yin, Mingyu Jin, Jin Pan +12
Test-time scaling improves language-model reasoning, but existing approaches often face a difficult trade-off: long chain-of-thought sampling remains single-threaded, while sentenc…
Learning Action Priors for Cross-embodiment Robot Manipulation
Dong Jing, Tianqi Zhang, Jiaqi Liu +5
Most Vision-Language-Action (VLA) models build on a Vision-Language Model (VLM) backbone by attaching an action module and optimizing the full policy jointly. This design inherits…
All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs
Xi Chen, Mingyu Jin, Jingcheng Niu +7
In this paper, we present empirical and theoretical evidence against a central but largely implicit assumption in circuit and sheaf discovery (CSD), which we term the Functional An…
Fin-Bias: Comprehensive Evaluation for LLM Decision-Making under human bias in Finance Domain
Xiaoyu Hu, Jinman Zhao
Large language models (LLMs) are increasingly deployed in financial contexts, raising critical concerns about reliability, alignment, and susceptibility to adversarial manipulation…
What Happens Inside Agent Memory? Circuit Analysis from Emergence to Diagnosis
Xutao Mao, Jinman Zhao, Gerald Penn +1
Agent memory failures are silent: an LLM-based agent can produce a fluent response even when it fails to extract, retain, or retrieve the information needed across sessions. The wr…
EndPrompt: Efficient Long-Context Extension via Terminal Anchoring
Han Tian, Luxuan Chen, Xinran Chen +10
Extending the context window of large language models typically requires training on sequences at the target length, incurring quadratic memory and computational costs that make lo…
Reinforcing Consistency in Video MLLMs with Structured Rewards
Yihao Quan, Zeru Shi, Jinman Zhao +1
Multimodal large language models (MLLMs) have achieved remarkable progress in video understanding. However, seemingly plausible outputs often suffer from poor visual and temporal g…
RealRoute: Dynamic Query Routing System via Retrieve-then-Verify Paradigm
Jiahe Liu, Qinkai Yu, Jingcheng Niu +5
Despite the success of Retrieval-Augmented Generation (RAG) in grounding LLMs with external knowledge, its application over heterogeneous sources (e.g., private databases, global c…
Not All Preferences Deserve Gradients: Understanding Gradient Utility in Offline Reasoning Alignment
Hui Wu, Hengyi Cai, Jinman Zhao +6
Offline preference optimization aligns reasoning models from fixed chosen--rejected pairs, yet standard methods apply gradient updates from every pair regardless of its training va…
Enhancing Video Representations with Spatiotemporal-Semantic Residual to Mitigate Hallucinations in Video Large Multimodal Models
Yuansheng Gao, Jinman Zhao, Tong Zhang +5
Although Video Large Multimodal Models have achieved strong performance in video understanding, they still suffer from hallucination. Existing inference-time intervention methods u…
-GRPO: Unifying the GRPO Frameworks with Learnable Token Preferences
Yining Wang, Jinman Zhao, Chuangxin Zhao +3
Reinforcement Learning with Human Feedback (RLHF) has been the dominant approach for improving the reasoning capabilities of Large Language Models (LLMs). Recently, Reinforcement L…
Listwise Direct Preference Optimization with Multi-Dimensional Preference Mixing
Yuhui Sun, Xiyao Wang, Zixi Li +6
Recent alignment methods based on Direct Preference Optimization (DPO) reformulate preference learning as supervised optimization over pairwise comparisons, offering improved effic…
Coarse-to-Fine Process Reward Modeling for Mathematical Reasoning
Yulan Hu, Sheng Ouyang, Jinman Zhao +1
The Process Reward Model (PRM) plays a crucial role in mathematical reasoning tasks, requiring high-quality supervised process data. However, we observe that reasoning steps genera…