13 papers
DynaCF: Mitigating Shortcut Learning in Reward Models via Dynamic Counterfactual Sensitivity
Fengyuan Liu, Yongliang Miao, Zirui He +3
Reward models trained from pairwise preferences often exploit superficial shortcut cues rather than learning true response quality. We propose DynaCF, a dynamic reweighting framewo…
SAEExplainer: Interpreting SAE Features with Activation-Guided Preference Optimization
Jingyi He, Haiyan Zhao, Ruxue Shi +4
Although Sparse Autoencoders (SAEs) have mitigated the opacity of large language models (LLMs) by decomposing dense representations into sparse features, explaining these features…
RASFT: Rollout-Adaptive Supervised Fine-Tuning for Reasoning
Yongliang Miao, Fengyuan Liu, Wei Shi +4
Supervised fine-tuning (SFT) is a prevailing method for adapting large language models to reasoning tasks by imitating offline expert demonstrations, often treating a single expert…
HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models
Shuang Liu, Yuxuan Bo, Qiuyang Zhao +4
Reward models are central to large language model (LLM) alignment, but they remain vulnerable to reward hacking. To evaluate reward-model robustness, we introduce RewardHackBench c…
Law of Neural Interaction: Depth-Width Shape, Interaction Efficiency, and Generalization
Wenjie Sun, Jinning Yang, Shuai Zhang +1
The guidance of scaling laws has increased the resource demands of modern large language models (LLMs), yet it remains questionable whether these models utilize resources effective…
Universal Activation Verbalizer: A Unified Framework for Cross-Model Activation Explanation
Haiyan Zhao, Zirui He, Guanchu Wang +3
Activation verbalization explains hidden representations in natural language, but existing methods are mostly limited to self-explanation, where each model explains only its own ac…