STRIVE: Structured Spatiotemporal Exploration for Reinforcement Learning in Video Question Answering
Emad Bahrami 1,4 Olga Zatsarynna 1,4 Parth Pathak 3 Sunando Sengupta 2 Juergen Gall 1,4 Mohsen Fayyaz 2 1 University of Bonn 2 Microsoft 3 Meta 4 Lamarr Institute for Machine Learning and Artificial Intelligence · 2026
Read STRIVE: Structured Spatiotemporal Exploration for Reinforcement Learning in Video Question Answering by Emad Bahrami 1,4 Olga Zatsarynna 1,4 Parth Pathak 3 Sunando Sengupta 2 Juergen Gall 1,4 Mohsen Fayyaz 2 1 University of Bonn 2 Microsoft 3 Meta 4 Lamarr Institute for Machine Learning and Artificial Intelligence (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-2604-01824.txt.
Description
Bahrami and colleagues present STRIVE, a reinforcement learning framework for video question answering. Group-based policy methods suffer low reward spread when answers are alike, giving weak advantage estimates. STRIVE builds several versions of each video and scales jointly across text outputs and visual versions, widening contrasts to visual changes. A sampling step favors frames most apt for the question while keeping time coverage. Tests on six video benchmarks show steady gains over strong rivals across large models.
Catalog record
| CREATOR | Emad Bahrami 1,4 Olga Zatsarynna 1,4 Parth Pathak 3 Sunando Sengupta 2 Juergen Gall 1,4 Mohsen Fayyaz 2 1 University of Bonn 2 Microsoft 3 Meta 4 Lamarr Institute for Machine Learning and Artificial Intelligence |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39074 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.01824