AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Model Collapse Is Not a Bug but a Feature in Machine Unlearning for LLMs

Yan Scholten, Sophie Xhonneux, Leo Schwinn, Stephan Günnemann

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 6 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 Partial Model Collapse unlearning instead of target-based baselinesPhi-1.5, Llama-3.2-3B-Instruct, Gemma-3-12b-it · TOFUutility vs unlearn quality Pareto frontImproved
replace NPO target-based unlearning with PMCLlama / Phi · Wikitext token probabilitiesprobability difference of forget tokens: -0.12 mean for NPO vs zero-mean for PMCImproved
replace baselines with PMC under sampling and prefilling attacksunlearned models · TOFU forget setworst-case ROUGE-L leakageImproved
train more epochs (up to 20) (ablation)PMC · TOFU (Phi-1.5)unlearn qualityImproved
increase number of samples (ablation)PMC · TOFU (Phi-1.5)unlearn qualityImproved
increase lambda trade-off (ablation)PMC · TOFU (Phi-1.5)utility and unlearn qualityMixed
increase sampling temperature (ablation)PMC · TOFU (Phi-1.5)unlearn quality and utilityMixed
use fewer than 10 samples (ablation)PMC · TOFU (Phi-1.5)collapse to max rewardHurt
use self-BLEU reward instead of ROUGE-L reward (ablation)PMC · TOFU (Llama-3.2-3B-Instruct)utilityImproved
add PMC fine-tuning (vs base) on general benchmarksPhi-1.5 · Arc-Easyaccuracy: 0.7622 -> 0.6891Hurt
add PMC fine-tuning (vs base) on general benchmarksLlama-3.2-3B-Instruct · MMLUaccuracy: 0.6041 -> 0.5924No clear effect
use gradient checkpointing (ablation)PMC · TOFUunlearning trajectoryHurt

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