CASI: Credit Assignment via Set Intervention for Agentic Retrieval

Key points

• Large language models increasingly generate multiple retrieval actions jointly, while downstream systems often provide only one reward for the resulting set.1
• Broadcasting this reward provides a valid score-function gradient estimator for the expected set-level objective, but does not determine a coalition-dependent within-set attribution.1
• Under the training-time evaluator, CASI achieves the highest median cosine similarity to a replay-defined Shapley reference among evaluated credit rules; under a sensitivity backend, CASI does not outperform Singleton.1
• Across five independent training runs, CASI-trained policies achieve higher held-out auction-replay set value than Singleton- or Broadcast-trained policies.1

Sources (1)

• [1]CASI: Credit Assignment via Set Intervention for Agentic Retrieval
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 02:47 PM

  We introduce CASI (Credit Assignment via Set Intervention for Agentic Retrieval), which constructs the missing counterfactual information through snapshot-grounded set intervention: generate and retrieve once, hold replayable snapshot components fixed, and evaluate alternative subsets under common conditions.

  Large language models increasingly generate multiple retrieval actions jointly, while downstream systems often provide only one reward for the resulting set.

Extractive summary: sentences quoted from the sources.