aiqbench analysis
This study reveals that sample retention during fine-tuning is reproducible and architecture-specific, contradicting the assumption that forgetting patterns are inherent properties of individual data points. The research demonstrates that standard models like Ebbinghaus exponential decay fail to capture these dynamics, showing instead complex non-monotonic behaviors across different datasets and backbones. Consequently, popular techniques relying on stable sample difficulty scores or uniform sampling may not yield significant generalization improvements as previously hypothesized.
Not All Forgetting Is Equal: Retention Dynamics in Fine-Tuned Image Classifiers
arXiv:2604.11508v3 Announce Type: replace Abstract: Fine-tuning a pretrained classifier leaves some samples reliably learned and others cycling between correct and incorrect. Curriculum learning, data pruning and dataset cartography assume that pattern is a property of the sample, untested. We record per-sample correctness at every epoch while fine-tuning ResNet-18 and DeiT-Small on an imbalanced retinal OCT dataset and CUB-200-2011, matching samples by image identity and holding the split fixed across seeds. Per-sample retention is reproducible: cross-run Spearman correlation of the fitted decay constant is 0.37 to 0.59 over ten seeds. It is architecture-specific: two runs of one backbone agree more than two backbones on identical data (0.45 and 0.59 within against 0.30 between on OCTDL). Loss after five frozen-backbone epochs predicts a different run's decay constant at 0.29 to 0.43. The Ebbinghaus exponential does not survive: monotone decay, the one shape it can represent, is 0.1% to 0.8% of samples, and on traces that do forget mean R-squared is negative in all four configurations. A power law and a free-asymptote variant fail on the same traces: the defect is monotonicity. Across five sampling arms with matched exposure, prioritisation ratios of 2.7x to 28x, and an online variant, none of 48 comparisons against uniform sampling survives Benjamini-Hochberg correction, though three seeds detect only about four accuracy points. A stable, cheap difficulty score does not buy generalisation through sampling. Patient-grouped splitting, the remedy for a leak reaching 76% to 78% of OCT test images, moves that dataset's headline metrics by less than their run-to-run spread.
Abstract:Fine-tuning a pretrained classifier leaves some samples reliably learned and others cycling between correct and incorrect. Curriculum learning, data pruning and dataset cartography assume that pattern is a property of the sample, untested. We record per-sample correctness at every epoch while fine-tuning ResNet-18 and DeiT-Small on an imbalanced retinal OCT dataset and CUB-200-2011, matching samples by image identity and holding the split fixed across seeds. Per-sample retention is reproducible: cross-run Spearman correlation of the fitted decay constant is 0.37 to 0.59 over ten seeds. It is architecture-specific: two runs of one backbone agree more than two backbones on identical data (0.45 and 0.59 within against 0.30 between on OCTDL). Loss after five frozen-backbone epochs predicts a different run's decay constant at 0.29 to 0.43. The Ebbinghaus exponential does not survive: monotone decay, the one shape it can represent, is 0.1% to 0.8% of samples, and on traces that do forget mean R-squared is negative in all four configurations. A power law and a free-asymptote variant fail on the same traces: the defect is monotonicity. Across five sampling arms with matched exposure, prioritisation ratios of 2.7x to 28x, and an online variant, none of 48 comparisons against uniform sampling survives Benjamini-Hochberg correction, though three seeds detect only about four accuracy points. A stable, cheap difficulty score does not buy generalisation through sampling. Patient-grouped splitting, the remedy for a leak reaching 76% to 78% of OCT test images, moves that dataset's headline metrics by less than their run-to-run spread.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2604.11508 [cs.LG]
(or arXiv:2604.11508v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2604.11508 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 342 words
Related coverage
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
Exploring self-distilled reasoning for supervised fine-tuning with Amazon Nova
AWS · Jul 21, 2026