Best-of-Both-Worlds for Heavy-Tailed Markov Decision Processes
Yu Chen, Yuhao Liu, Jiatai Huang, Yihan Du, Longbo Huang · 2026
Read Best-of-Both-Worlds for Heavy-Tailed Markov Decision Processes by Yu Chen, Yuhao Liu, Jiatai Huang, Yihan Du, Longbo Huang (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-2602-01295.txt.
Description
Yu Chen and co-authors build best-of-both plans for Markov tasks with heavy-tail loss. Past heavy-tail plans play too safe in calm worlds and fail to bend in hostile ones. Their HT-FTRL plans use follow-the-leader math over occupancy maps with skip-loss guesses, earning flat regret in hostile worlds and log regret in calm worlds. A tougher variant tackles unknown transitions with a gloomy skip guess under a mild rule. Key tricks tame heavy-tail shifts and split transition doubt from loss skew.
Catalog record
| CREATOR | Yu Chen, Yuhao Liu, Jiatai Huang, Yihan Du, Longbo Huang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 156237 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01295