{"paper":{"title":"Efficient Multi-objective Prompt Optimization via Pure-exploration Bandits","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Multi-objective prompt selection for large language models reduces to pure-exploration bandit problems, enabling efficient algorithms with theoretical guarantees.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chengshuai Shi, Cong Shen, Donghao Li, Jing Yang, Weijuan Ou","submitted_at":"2026-05-14T08:31:17Z","abstract_excerpt":"Prompt engineering has become central to eliciting the capabilities of large language models (LLMs). At its core lies prompt selection -- efficiently identifying the most effective prompts. However, most prior investigations overlook a key challenge: the inherently multi-faceted nature of prompt performance, which cannot be captured by a single metric. To fill this gap, we study the multi-objective prompt selection problem under two practical settings: Pareto prompt set recovery and best feasible prompt identification. Casting the problem into the pure-exploration bandits framework, we adapt p"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Casting the problem into the pure-exploration bandits framework, we adapt provably efficient algorithms from multi-objective bandits and further introduce a novel design for best feasible arm identification in structured bandits, with theoretical guarantees on the identification error in the linear case. Extensive experiments across multiple LLMs show that the bandit-based approaches yield significant improvements over baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Prompt performance across multiple objectives can be modeled as rewards from independent or linearly structured arms in a pure-exploration bandit setting without significant interference or non-stationarity from LLM stochasticity.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Adapting multi-objective pure-exploration bandits enables efficient Pareto prompt set recovery and best feasible prompt identification for LLMs, with linear-case guarantees and empirical gains over baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Multi-objective prompt selection for large language models reduces to pure-exploration bandit problems, enabling efficient algorithms with theoretical guarantees.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c558920ac2ac792c7247d4fc736c4717fb47f4e8affad22e63b6cedd3eccd3f1"},"source":{"id":"2605.14553","kind":"arxiv","version":1},"verdict":{"id":"ae9d7599-fe60-4212-afc9-226b40e7a23c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:47:29.025982Z","strongest_claim":"Casting the problem into the pure-exploration bandits framework, we adapt provably efficient algorithms from multi-objective bandits and further introduce a novel design for best feasible arm identification in structured bandits, with theoretical guarantees on the identification error in the linear case. Extensive experiments across multiple LLMs show that the bandit-based approaches yield significant improvements over baselines.","one_line_summary":"Adapting multi-objective pure-exploration bandits enables efficient Pareto prompt set recovery and best feasible prompt identification for LLMs, with linear-case guarantees and empirical gains over baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Prompt performance across multiple objectives can be modeled as rewards from independent or linearly structured arms in a pure-exploration bandit setting without significant interference or non-stationarity from LLM stochasticity.","pith_extraction_headline":"Multi-objective prompt selection for large language models reduces to pure-exploration bandit problems, enabling efficient algorithms with theoretical guarantees."},"references":{"count":33,"sample":[{"doi":"","year":2025,"title":"Gepa: Reflective prompt evolution can outperform reinforcement learning","work_id":"3de5a529-96e4-4800-82cd-546562b68f98","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2010,"title":"Best arm identification in multi-armed bandits","work_id":"8c40c1c5-f69c-4f4d-966c-e335bdb85b4e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1901,"title":"Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901","work_id":"84780adb-76cf-44ac-a8b7-e24d4fa5c592","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Leaf: A benchmark for federated settings","work_id":"94ec8be1-6cf1-4962-994c-e762a0b9c3ee","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Discrete prompt optimization via constrained generation for zero-shot re-ranker","work_id":"0c01ab93-25b2-4bca-b51b-919144ac6c2c","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":33,"snapshot_sha256":"9a0161828a6a59d6c66ffc98eb97c27ce3a8e511285d867ad587ec3324b600cd","internal_anchors":4},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}