Mitigating catastrophic forgetting in lifelong learning: a hybrid architecture integrating neural ordinary differential equations with memory-augmented transformers
Song Zhou, Qiang Li · 2025
Read Mitigating catastrophic forgetting in lifelong learning: a hybrid architecture integrating neural ordinary differential equations with memory-augmented transformers by Song Zhou, Qiang Li (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/pmc-12808139.txt.
Description
A net that learns task after task tends to wipe the first as it takes the next, a flaw named for its harsh toll. Song Zhou and Qiang Li wed two builds to fight that wipe in a 2025 study. One build flows in smooth time steps; the twin stacks a mind bank that the net can call back by aim.
The pair add strict learn bounds that tie bank size, task chain length, and wipe depth, plus a smart rule for what the bank keeps. On three hard test sets, the wed build cut the wipe by near a fourth and beat top past builds on truth by a tenth. The bounds give crews a scale for such builds.
Catalog record
| CREATOR | Song Zhou, Qiang Li |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |