AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Annotations Mitigate Post-Training Mode Collapse

Jacob Mitchell Springer, Madhu Advani, Lukas Aichberger, Arwen Bradley, Eran Malach, Omid Saremi, Sinead Williamson, Preetum Nakkiran, Etai Littwin, Aditi Raghunathan

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining · 0 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
annotation-anchored training instead of standard SFT2.5B model pretrained and post-trained on Tulu3 · STORIESsemantic entropy: 6.88 -> 8.30Improved
add annotated pretraining before annotation-anchored post-training (ablation)annotation-anchored 2.5B model · STORIESsemantic entropy: 7.07 -> 8.30Improved
sample annotations at inference time (ablation)annotation-anchored 2.5B model · STORIESsemantic entropy: 7.01 -> 8.30Improved
increase model size under standard SFTQwen2.5 and Llama 3 instruction-tuned models · STORIESsemantic entropyHurt
increase number of SFT post-training examples (18K to 464K)standard PT + SFT · STORIESsemantic entropyHurt
prompt for diversity (brainstorm, multiple responses) instead of direct promptinginstruction-tuned Qwen2.5 / Llama 3 · STORIESsemantic entropyMixed
use annotation-anchored training instead of standard SFT2.5B model · GSM8Kaccuracy: 35.4% -> 36.4%No clear effect
use annotation-anchored training instead of standard SFT0.6B, 1B, 2.5B models · Zero-shot suite averageaverage accuracy: 62.3 vs 62.4 at 2.5BNo clear effect
use annotation-anchored training instead of SFT across temperaturesstandard PT + SFT · STORIES diversity-qualitydiversity at given qualityImproved
narrow post-training data entropy, with annotation anchoring vs standard SFT2.5B models on SimpleStories · SimpleStories topic/persona entropyoutput entropyImproved

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