Failure is Feedback : History-Aware Backtracking for Agentic Traversal in Multimodal Graphs
Joohyung Yun, Doyup Lee, Wook-Shin Han · 2026
Read Failure is Feedback : History-Aware Backtracking for Agentic Traversal in Multimodal Graphs by Joohyung Yun, Doyup Lee, Wook-Shin Han (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-2602-03432.txt.
Description
Searching a web of linked documents for one paragraph, table, or image is the job of multimodal retrieval. Joohyung Yun and 2 co-authors criticize graph-based retrievers in a 2026 paper for two habits. They score every hop the same way, and their fixed plans cannot recover from a wrong turn. Their system is called Failure is Feedback, or FiF.
FiF treats retrieval as a sequence of decisions made by an agent. When a path fails, the system backtracks but carries the lessons of the failure with it, instead of simply resetting. A cost-aware controller spends cheap search first and calls in heavy language-model reasoning only when earlier failure justifies the cost. FiF tops three multimodal question-answering benchmarks.
Catalog record
| CREATOR | Joohyung Yun, Doyup Lee, Wook-Shin Han |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 76 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03432