14 papers
MCPHallu: Benchmarking Reasoning, Execution, and Memory Hallucinations in MCP Agents
Yizhen Jiang, Haowei Guo, Tianyi Bai +6
Large language model agents are increasingly deployed through the Model Context Protocol (MCP), which allows them to discover and invoke tools across multiple servers at runtime. E…
Compile, Don't Memorize: A Context Compilation Architecture (CCA) for In-Context Learning
Jinhu Qi, Minda Hu, Wentao Zhang +4
Large language models (LLMs) increasingly handle in-context learning (ICL) tasks where a long, novel context defines the rules, knowledge, and output schema for a series of questio…
TREK: A Travel Reasoning and Evaluation Kit for LLM Agents in Complex Trip Planning
Jinhu Qi, Wentao Zhang, Siu Man Ng +4
Travel planning is a demanding stress test for tool-using LLM agents: a usable itinerary is a single artifact that must be right along many axes at once - every flight, hotel, and…
Geometric Collapse: When Vision Models Fail to Verify Physical Causality
Wentao Zhang, Jinhu Qi, Weiqiang Jin +3
Recent progress in large-scale self-supervised learning has improved dense geometric prediction, but it remains unclear whether such scaling yields inference-time physical plausibi…
When Should the Teacher Move? Temporal Coupling and Stability in Self On-Policy Distillation
Haowei Guo, Baolong Bi, Ruicheng Zhang +4
Self on-policy distillation trains a student policy against a teacher derived from its own parameter history, yet the teacher's update schedule -- which governs the \emph{temporal…
One-Shot Klein Cutting Planes for Lipschitz Geodesically Convex Optimization in Hyperbolic Space
Yutong Zhang, Yaoran Yang, Yifan Zhu +1
Motivated by the COLT 2023 open problem of Criscitiello, Martínez-Rubio, and Boumal on deterministic first-order methods for Lipschitz geodesically convex optimization on Hadamard…
TRACER: Turn-level Regret Matching with Inner Reinforcement Credit for Cooperative Multi-LLM Reasoning
Chusen Li, Zhou Liu, Shuigeng Zhou +1
Large language models increasingly rely on either reinforcement learning or multi-agent prompting to improve reasoning, yet these two paradigms remain difficult to combine. Directl…
LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning
Yifan Dai, Zhenhua Wu, Bohan Zeng +18
Joint audio-visual reasoning is essential for omnimodal understanding, yet current multimodal large language models (MLLMs) still struggle when reasoning requires fine-grained evid…
Towards Next-Generation LLM Training: From the Data-Centric Perspective
Hao Liang, Zhengyang Zhao, Zhaoyang Han +8
Large language models (LLMs) have demonstrated remarkable performance across a wide range of tasks and domains, with data playing a central role in enabling these advances. Despite…
LongVidSearch: An Agentic Benchmark for Multi-hop Evidence Retrieval Planning in Long Videos
Rongyi Yu, Chenyuan Duan, Wentao Zhang
Long video question answering (Long-Video QA) increasingly relies on agentic tool use to retrieve evidence from long videos. In realistic settings, this process often requires mult…
Beyond Benchmark Islands: Toward Representative Trustworthiness Evaluation for Agentic AI
Jinhu Qi, Yifan Li, Minghao Zhao +4
Agentic AI systems increasingly act through tool-augmented, multi-step workflows whose failures (unsafe tool use, unauthorised actions, social harm) carry deployment-level conseque…
GIFT: Reconciling Post-Training Objectives via Variational Finite-Temperature Gibbs Initialization
Zhengyang Zhao, Lu Ma, Yizhen Jiang +7
The prevailing post-training paradigm for Large Reasoning Models (LRMs)---Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL)-suffers from an intrinsic optimizatio…
Leash: Adaptive Length Penalty and Reward Shaping for Efficient Large Reasoning Model
Yanhao Li, Lu Ma, Jiaran Zhang +3
Existing approaches typically rely on fixed length penalties, but such penalties are hard to tune and fail to adapt to the evolving reasoning abilities of LLMs, leading to suboptim…
Overcoming Environmental Meta-Stationarity in MARL via Adaptive Curriculum and Counterfactual Group Advantage
Weiqiang Jin, Yang Liu, Shixiang Tang +5
Multi-agent reinforcement learning (MARL) has reached competitive performance on cooperative tasks against scripted adversaries, yet most methods train agents at a single fixed dif…