Model Inversion with Layer-Specific Modeling and Alignment for Data-Free Continual Learning
Ruilin Tong, Haodong Lu, Yuhang Liu, Dong Gong · 2025
Read Model Inversion with Layer-Specific Modeling and Alignment for Data-Free Continual Learning by Ruilin Tong, Haodong Lu, Yuhang Liu, Dong Gong (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-2510-26311.txt.
Description
Ruilin Tong, Haodong Lu, Yuhang Liu, and Dong Gong improve data free continual learning by synthesizing old data from the model itself. Continual learning trains a model on a sequence of tasks without forgetting earlier ones, but storing old data for replay is often barred by privacy or security. Model inversion generates synthetic inputs from a trained model, yet generating from compressed output labels alone causes drift from real data, and full model inversion is expensive.
Their per layer model inversion optimizes one layer at a time. The idea comes from faster convergence in single layer problems. It provides a strong starting point for full inversion and cuts iterations sharply. Layer specific modeling and alignment then reduce the drift between synthetic and real data, which matters most for large pretrained models.
Catalog record
| CREATOR | Ruilin Tong, Haodong Lu, Yuhang Liu, Dong Gong |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 98 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-26311