Xiaocheng Feng

Unlocking Multilingual Reasoning Capabilities of LLMs and LVLMs via Representation Engineering

CRISP: Cultural Reward Modeling for Implicit Situated Propriety

A culturally situated reward model that assigns rewards according to cultural appropriateness in open-ended social scenarios, and a collaborative multi-agent framework that instantiates implicit cultural norms into diverse social scenarios to construct culturally situated data are introduced.

Rethinking Thinking Steps in Overthinking: Towards More Effective Reasoning

This work proposes a training-free, interpretable framework that selects thinking words via attention heads to guide LRMs toward more effective reasoning, and consistently outperforms the strong baseline DEER.

ACTD: Anchor-Based Cross-Tokenizer Distillation with Residual Regularization

An Anchor-Based Cross-Tokenizer Distillation with Residual Regularization with Residual Regularization (ACTD) bridges structural heterogeneity through vocabulary and sequence alignment, while mitigating alignment noise via a novel anchor loss with residual regularization.

H2SD: Hybrid Hindsight Self-Distillation

Experiments on challenging reasoning benchmarks show that H$^2$SD achieves the strongest overall performance among representative RLVR and self-distillation baselines, with stable optimization and a favorable accuracy-efficiency trade-off.