RAVEL Learns Retrieval-Aware Questioning for Interactive Person Re-identification
What Should We Ask Next? Retrieval-Aware Question Learning under Partial Evidence
The framework replaces imitation of offline QA orderings with online rank feedback, giving practitioners a concrete recipe for training question-asking retrieval agents on benchmarks like Interactive-PEDES.
Researchers introduce RAVEL, an online reinforcement learning framework that trains an agent to choose which question will produce the most useful evidence for the next retrieval update in interactive person re-identification.
Unlike prior systems that imitate an offline ordering of candidate QA pairs, RAVEL initializes from supervised question generation, observes the current Top-4 candidates, and optimizes its question policy with rank feedback from the full question-answer-retrieval loop.
On the Interactive-PEDES benchmark, RAVEL delivers progressively stronger retrieval performance across five interaction rounds.
The authors also show RAVEL reallocates its questioning budget toward localized open-ended attributes, which yield the largest gains on initially difficult queries.
What Should We Ask Next? Retrieval-Aware Question Learning under Partial Evidence
Interactive retrieval under partial evidence is a sequential information-acquisition problem: an agent must decide which question will create the most useful evidence for the next retrieval update. Existing systems train this decision by imitating an offline ordering of candidate QA pairs, although question value is determined by the response it elicits and its downstream effect on retrieval. We establish that candidate discriminativeness and perceived usefulness provide weak supervision for this objective, then introduce RAVEL, a retrieval-aware online reinforcement learning framework for interactive person re-identification. RAVEL initializes from supervised question generation, observes the current Top-4 candidates directly, and optimizes the question policy with rank feedback from the full question-answer-retrieval loop. Experiments on Interactive-PEDES show that RAVEL delivers progressively stronger retrieval performance across five interaction rounds. Further analysis shows that RAVEL reallocates the questioning budget toward localized open-ended attributes, which provide more useful retrieval evidence and yield the largest gains on initially difficult queries.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include RAVEL, Interactive-PEDES, reinforcement learning, person re-identification.
Source attribution: curated from arXiv cs.AI (@Lyucheng Qian, John Yuehan Zhang, Pingyu Wang).