Prototype-Based Continual Learning with Label-free Replay Buffer and Cluster Preservation Loss
Agil Aghasanli, Yi Li, Plamen Angelov · 2025
Read Prototype-Based Continual Learning with Label-free Replay Buffer and Cluster Preservation Loss by Agil Aghasanli, Yi Li, Plamen Angelov (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-07240.txt.
Description
Agil Aghasanli, Yi Li, and Plamen Angelov built a learning system that keeps old skills while it gains new ones, without labels on its stored memory, in a paper published in 2025. Most such systems store past samples with their labels and replay them. Their store holds bare prototypes, the core example of each past class, kept with no label. A new loss function guards the shape of old groups in the model's internal space.
Two mechanisms shape the stored prototypes: one pulls new learning toward them, and one pushes clashing cases away. The scheme fits both new class rounds and new domain rounds. On six benchmark datasets, built on strong pretrained features, it beat the best known continual learning methods. In places it even beat training on all data at once. A fully unlabeled version also matched some label-fed rivals.
Catalog record
| CREATOR | Agil Aghasanli, Yi Li, Plamen Angelov |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-07240