Mutual Information Surprise: Rethinking Unexpectedness in Autonomous Systems
Yinsong Wang, Quan Zeng, Xiao Liu, Yu Ding, H. Milton Stewart School of Industrial, Systems Engineering, Georgia Institute of Technology, Atlanta & 30332, USA · 2025
Read Mutual Information Surprise: Rethinking Unexpectedness in Autonomous Systems by Yinsong Wang, Quan Zeng, Xiao Liu, Yu Ding, H. Milton Stewart School of Industrial, Systems Engineering, Georgia Institute of Technology, Atlanta & 30332, USA (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-2508-17403.txt.
Description
Engineers often give machines a sense of surprise, measured by how unlikely or belief-changing an event is. Yinsong Wang and co-authors argue that surprise should mean something else: a sign that the system has grown in knowledge. Their measure, mutual information surprise, gauges how much a new observation changes what the system's parts share with each other — its learned structure — rather than how strange the observation looks.
They add a statistical test sequence that can trigger a surprise reaction, plus a response policy that adjusts sampling and forks the process when surprise strikes. On synthetic tasks and on a live pollution-mapping problem, systems run by their policy beat classical surprise-based ones in stability, responsiveness, and accuracy. The authors cast the shift as moving surprise from reaction to reflection, a step toward machines that monitor their own learning and adapt with awareness.
Catalog record
| CREATOR | Yinsong Wang, Quan Zeng, Xiao Liu, Yu Ding, H. Milton Stewart School of Industrial, Systems Engineering, Georgia Institute of Technology, Atlanta & 30332, USA |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-17403