Forget Less by Learning from Parents Through Hierarchical Relationships
Arjun Ramesh Kaushik, Naresh Kumar Devulapally, Vishnu Suresh Lokhande, Nalini K. Ratha, Venu Govindaraju · 2026
Read Forget Less by Learning from Parents Through Hierarchical Relationships by Arjun Ramesh Kaushik, Naresh Kumar Devulapally, Vishnu Suresh Lokhande, Nalini K. Ratha, Venu Govindaraju (2026) 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-2601-01892.txt.
Description
Personalized image generators learn one concept after another, but each new concept tends to erase earlier ones, a failure called catastrophic forgetting. Arjun Ramesh Kaushik and four co-authors propose learning from parents: older concepts actively guide the learning of new ones through relationships arranged in hyperbolic space.
Their framework, FLLP, embeds concepts on a Lorentzian manifold whose geometry naturally fits tree-like hierarchies, so parent concepts shelter their children while new ones are integrated. Tests on three public datasets and one synthetic benchmark show consistent gains in robustness and generalization over methods that merely reduce interference between concepts. Treating prior knowledge as a teacher rather than a rival is the paper's central idea.
Catalog record
| CREATOR | Arjun Ramesh Kaushik, Naresh Kumar Devulapally, Vishnu Suresh Lokhande, Nalini K. Ratha, Venu Govindaraju |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-01892