Research
Annotation Entropy Predicts Per-Example Learning Dynamics in LoRA Fine-Tuning
arXiv:2604.16332v1 Announce Type: cross Abstract: We find that LoRA fine-tuning exhibits un-learning on contested examples: items with high annotator disagreement show increasing loss during training,
arXiv:2604.16332v1 Announce Type: cross Abstract: We find that LoRA fine-tuning exhibits un-learning on contested examples: items with high annotator disagreement show increasing loss during training, a qualitatively distinct pattern largely absent under full fine-tuning and consistent across all six models tested (four encoder, two decoder-only). This discovery emerges from correlating annotation entropy, computed from ChaosNLI's 100 labels per example, with per-example area under the loss curve (AULC) on SNLI and MNLI. The correlation is positive in all 25 conditions tested (Spearman rho = 0.06-0.43), with decoder-only models showing stronger correlations than encoders at matched LoRA rank. The effect survives partial-correlation controls and replicates across seeds and datasets. A preliminary noise-injection experiment is consistent with these findings.
Related
- (How) Learning Rates Regulate Catastrophic Overtraining
- Testing the Assumptions of Active Learning for Translation Tasks with Few Samples
- Detecting and refurbishing ground truth errors during training of deep learning-based echocardiography segmentation models
- Enhancing Trust in Large Language Models via Uncertainty-Calibrated Fine-Tuning
- Beyond Black-Box Labels: Interpretable Criteria for Diagnosing SubjectiveNLP Tasks
Source: arXiv cs.CL | 2026-04-21