AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

ASKD: Reinforcement Learning-Style Knowledge Distillation with Quality-Adaptive Skewness

Mingjie Zhang, Xiaoling Zhou, Yuxiao Luo, Yiyu Liu, Shikun Zhang, Wei Ye

AAAI 2026 · Label-efficient and robust learning › Knowledge distillation · approach: Distillation with causal and adaptive guidance · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use ASKD instead of DistiLLM-2student distillation · Instruction following avg, Qwen2-1.5Bwin rate: 58.7 -> 60.0Improved
use ASKD instead of DistiLLM-2student distillation · Instruction following avg, Danube2-1.8Bwin rate: 56.5 -> 58.6Improved
use ASKD instead of DistiLLM-2student distillation · Instruction following avg, Gemma2-2Bwin rate: 74.8 -> 75.1Improved
use ASKD instead of DistiLLM-2Gemma2-2B · AlpacaEvalwin rate: 86.0 vs 84.2Hurt
use ASKD instead of DistiLLM-2Qwen2-Math-1.5B · GSM8K/MATH avgpass@1: 60.73 -> 62.06Improved
use ASKD instead of DistiLLM-2DS-Coder-1.3B · MBPPpass@1: 66.14 vs 66.14No clear effect
add adaptive skewness from quality rewards (ablation)ASKD · Math reasoning ablationavg score: 80.04 -> 81.90Improved
add Monte-Carlo RL-style S(R)KL formulation (ablation)ASKD · Math reasoning ablationavg score: 81.50 -> 81.90Improved
add adaptive skewness and high-quality SGO clipping (ablation)ASKD · Math reasoning ablationavg score: 80.66 -> 81.90Improved
use all three ASKD components vs none (ablation)ASKD · Math reasoning ablationavg score: 79.55 -> 81.90Improved
project advantage onto skewness instead of multiplying as in GRPO (ablation)SKL distillation · GSM8K/MATH, Qwen2-Math-1.5Bpass@1 avg: 61.42 -> 62.06Improved
project advantage onto skewness instead of multiplying as in GRPO (ablation)SKL distillation · GSM8K/MATH, Qwen2.5-Math-1.5Bpass@1 avg: 64.20 -> 64.80Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy