Transfer Learning in Latent Contextual Bandits with Covariate Shift Through Causal Transportability
Mingwei Deng, Ville Kyrki, Dominik Baumann · 2025
Read Transfer Learning in Latent Contextual Bandits with Covariate Shift Through Causal Transportability by Mingwei Deng, Ville Kyrki, Dominik Baumann (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-2502-20153.txt.
Description
A system that learned in one setting can stumble when moved to another, and copying all its knowledge over can even make things worse. Mingwei Deng, Ville Kyrki, and Dominik Baumann study that risk in this 2025 paper on bandit problems, where a learner picks actions and learns from rewards. The true context behind each choice is hidden, and only a rich stand-in signal can be seen.
The authors show that plain transfer of all knowledge does in fact hurt in this setup. Drawing on causal transport theory, they build methods that move only the knowledge that still holds in the new setting, where the context mix has shifted. A neural tool that compresses the stand-in signal helps estimate effects. On test data, the new methods learned faster than standard ones across several stand-ins.
Catalog record
| CREATOR | Mingwei Deng, Ville Kyrki, Dominik Baumann |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-20153