AI Paper Atlas

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

LLM-Oriented Token-Adaptive Knowledge Distillation

Xurong Xie, Zhucun Xue, Jiafu Wu, Jian Li, Yabiao Wang, Xiaobin Hu, Yong Liu, Jiangning Zhang

AAAI 2026 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 2 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
add AdaKD (LATF+IDTS) to FKDFKD · 5 benchmarks avg (Qwen2-7B->1.5B)ROUGE-L: 26.56 -> 26.63Improved
add AdaKD to RKDRKD · 5 benchmarks avg (Qwen2-7B->1.5B)ROUGE-L: 31.70 -> 32.97Improved
add AdaKD to GKDGKD · 5 benchmarks avg (OpenLLaMA2-7B->3B)ROUGE-L: 27.40 -> 29.28Improved
add AdaKD to ABKDABKD · 5 benchmarks avg (OpenLLaMA2-7B->3B)ROUGE-L: 28.95 -> 29.32Improved
add IDTS only to RKD (ablation)RKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 37.03 -> 38.55Improved
add LATF only to RKD (ablation)RKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 37.03 -> 37.07No clear effect
add LATF on top of IDTS (ablation)AdaKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 38.55 -> 39.01Improved
use Hellinger distance as difficulty indicator instead of alternatives (ablation)AdaKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 39.01 vs 38.38-38.85Mixed
use adaptive LATF instead of fixed r = 0.75 (ablation)AdaKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 38.59 -> 39.01Improved
use cosine schedule r 1.0->0.75 instead of LATF (ablation)AdaKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 39.01 -> 38.87No clear effect
reverse temperature direction (high T for hard tokens) (ablation)AdaKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 39.01 -> 35.67Hurt
use token-level IDTS instead of global fixed T=e^-0.5 (ablation)AdaKD · Dolly/S-NI/UnNI avg (Qwen2-7B->1.5B)ROUGE-L: 38.15 -> 39.01Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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