Causal Imitation Learning Under Measurement Error and Distribution Shift
Shi Bo, AmirEmad Ghassami · 2026
Read Causal Imitation Learning Under Measurement Error and Distribution Shift by Shi Bo, AmirEmad Ghassami (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-2601-22206.txt.
Description
Shi Bo and AmirEmad Ghassami study imitation learning when the student cannot see the world cleanly. In offline imitation, an agent learns from recorded expert behavior. Trouble starts when part of the state that drove the expert's decisions is observed only through noisy measurements, and the distribution shifts before deployment.
Those flaws create spurious links between states and actions. Standard behavioral cloning then converges to systematically wrong behavior, whether it uses the raw measurements or ignores them. The authors' 2026 preprint develops causal imitation learning that corrects for measurement error and distribution shift. The problem is practical: robots and medical policies alike learn from sensors that never report the true state.
Catalog record
| CREATOR | Shi Bo, AmirEmad Ghassami |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 82 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-22206