Metraipolis › archive › dc_cba49dbad267
What Must Replay Preserve? Separating Correctable Bias from Class Correspondence — arXiv:2610.07077
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.07077
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T11:40:49.566Z
- SHA-256 of the archived text
d2f5fc73602a95cca939636dd95a6b403302e4e046983bdd0fb18aaf7ef847b9- Stable URL of this page
https://metraipolis.com/documento/dc_cba49dbad267
How to cite
What Must Replay Preserve? Separating Correctable Bias from Class Correspondence — arXiv:2610.07077 — arXiv (en), https://arxiv.org/abs/2610.07077 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cba49dbad267 (Metraipolis archive dc_cba49dbad267, sha256 d2f5fc73602a95cca939636dd95a6b403302e4e046983bdd0fb18aaf7ef847b9)
BibTeX
@misc{dccba49dbad267,
title = {What Must Replay Preserve? Separating Correctable Bias from Class Correspondence — arXiv:2610.07077},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.07077},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cba49dbad267; sha256 d2f5fc73602a95cca939636dd95a6b403302e4e046983bdd0fb18aaf7ef847b9},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.07077. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Class-incremental learning must recognize all classes seen so far without task labels. Logit replay methods such as DER and DER++ mitigate forgetting by matching the model's past predictions on stored examples. Deleting this matching reveals its benefit, but the resulting accuracy cost cannot show whether the stored scores themselves are needed, or whether the cost survives correction of the classifier's bias toward recent classes. We propose a diagnostic framework that treats a cached prediction as temporally heterogeneous supervision: it separates classes known when an example was stored from classes learned afterward, edits each group, and evaluates every model before and after a task-level offset that leaves within-task predictions unchanged. On CIFAR-100 with DER++, suitable fixed constants replace the unrefreshed stored scores of later-learned classes within an equivalence margin of 1 percentage point, and the offset reduces the cost of deleting their matching from 14.9 to 1.8 points. Reassigning the non-gold scores of classes known at storage, which preserves their values and each task's target probability, costs 4.3 points before and 4.0 after the offset, and a parallel cost persists in image distillation. In the tested fixed-head setting, the large cost of deleting later-class matching is thus mostly correctable by this offset, whereas the smaller cost of disrupting class correspondence persists. Code and data are available at this http URL . Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.07077 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T11:19:24.724Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).