18 papers
AgentStream: How Well Do Self-Evolving LLM Agents Perform Under Streaming Tasks?
Dong Yan, Jian Liang, Dapeng Hu +4
Large language model (LLM) agents can self-evolve by continually improving from their own accumulated experience. However, existing studies predominantly adopt independent evaluati…
USE: A Unified Self-Ensembling Framework for Test-Time Prompt Tuning
Siru Jiang, Jian Liang, Ran He +1
Test-time adaptation (TTA) has emerged as a popular paradigm for improving the performance of vision-language models (e.g., CLIP) on downstream tasks. Among existing CLIP-based TTA…
On the Vulnerability of Parameter-Level Defenses to Model Merging
Kuangpu Guo, Qingyan Zheng, Jian Liang +4
The training-free integration of expert models via model merging has exposed significant security risks, enabling free-riders to combine specialized models without authorization. R…
Mitigating the Safety-utility Trade-off in LLM Alignment via Adaptive Safe Context Learning
Yanbo Wang, Minzheng Wang, Jian Liang +3
While reasoning models have achieved remarkable success in complex reasoning tasks, their increasing power necessitates stringent safety measures. For safety alignment, the core ch…
WorldCoder-Bench: Benchmarking Physically Grounded 3D World Synthesis
Shuo Lu, Yinuo Xu, Kecheng Yu +8
Large language models (LLMs) are increasingly asked not only to write static interfaces, but to construct executable interactive worlds from natural language. Browser-native 3D, co…
Understanding and Mitigating Spurious Signal Amplification in Test-Time Reinforcement Learning for Math Reasoning
Yongcan Yu, Lingxiao He, Jian Liang +5
Test-time reinforcement learning (TTRL) always adapts models at inference time via pseudo-labeling, leaving it vulnerable to spurious optimization signals from label noise. Through…