Institution

University of Oxford

One of the world's oldest universities, with machine learning research spanning deep learning theory, NLP, and reinforcement learning.

LLM Reasoning · Samsung Research

TrOPD: Trust-Region On-Policy Distillation for Small LLMs

TrOPD masks on-policy distillation to the tokens where the teacher is actually trustworthy, adding +3.06 to +3.52 average points over standard OPD on math, code, and STEM benchmarks with 1.5B-1.7B students.