Sign-Separated Finite-Time Error Analysis of Q-Learning
Donghwan Lee · 2026
Read Sign-Separated Finite-Time Error Analysis of Q-Learning by Donghwan Lee (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-2605-16103.txt.
Description
Lee analyzes Q-learning's overestimation bias, where picking the max of noisy value estimates inflates learned values above the true optima. Splitting the error into its plus and minus parts, he derives separate finite-time rates for each. Plus errors can carry a slower exponential envelope than minus errors, giving indirect evidence that max-picking amplifies plus errors. He also builds cases where the predicted gap shows up in real runs, with fixed-step bounds for both deterministic and stochastic settings.
Catalog record
| CREATOR | Donghwan Lee |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 89633 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.16103