Xiang Chen
7 papers · Latest:
ARIADNE: Agentic Reward-Informed Adaptive Decision Exploration via Blackboard-Driven MCTS for Competitive Program Generation
ARIADNE uses MCTS and a blackboard to significantly improve LLM-based competitive program generation by systematic exploration and feedback.
Unpaired Image Deraining Using Reward-Guided Self-Reinforcement Strategy
This paper introduces RGSUD, an unsupervised image deraining method that uses a reward-guided self-reinforcement strategy to leverage high-quality outputs during training.
From Mirage to Grounding: Towards Reliable Multimodal Circuit-to-Verilog Code Generation
New research reveals MLLMs often ignore visual input when generating code from circuit diagrams, proposing VeriGround to achieve genuine visual grounding.
DPN-LE: Dual Personality Neuron Localization and Editing for Large Language Models
DPN-LE precisely edits personality in LLMs by targeting specific, mutually exclusive neurons, preserving general capabilities better than prior methods.
Security Attack and Defense Strategies for Autonomous Agent Frameworks: A Layered Review with OpenClaw as a Case Study
This paper provides a layered review of security risks and defense strategies for LLM-based autonomous agent frameworks, using OpenClaw as a case study.
FreqCache: Accelerating Embodied VLN Models with Adaptive Frequency-Guided Token Caching
FreqCache accelerates Vision-Language-Navigation (VLN) models by using adaptive frequency-guided token caching, achieving 1.59x speedup.
Revealing Physical-World Semantic Vulnerabilities: Universal Adversarial Patches for Infrared Vision-Language Models
A new universal adversarial patch (UCGP) exposes critical semantic vulnerabilities in infrared vision-language models, compromising low-visibility perception.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.