When Do Early-Exit Networks Generalize? A PAC-Bayesian Theory of Adaptive Depth
Dongxin Guo 1 , Jikun Wu 2 , and Siu Ming Yiu 1 · 2026
Read When Do Early-Exit Networks Generalize? A PAC-Bayesian Theory of Adaptive Depth by Dongxin Guo 1 , Jikun Wu 2 , and Siu Ming Yiu 1 (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-2604-15764.txt.
Description
Guo, Wu, and Yiu build a PAC-Bayes frame for early-exit nets, where sure guesses leave at middle layers for faster asks. Their bounds lean on exit-depth spread and mean depth, not max depth, with cost of order mean depth times dim plus spread over epsilon squared. The frame proves when bendy-depth nets beat fixed-depth nets. Tests on 6 builds and 7 test beds fit within 1.52 to 3.87 times, far tighter than old bounds past 100 times.
Catalog record
| CREATOR | Dongxin Guo 1 , Jikun Wu 2 , and Siu Ming Yiu 1 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28532 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.15764