AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

DAPS: Dependency-Aware Premise Selection for LLM Theorem Proving

Zixuan Chen, Wenyuan Jiang, Junling Wang, Mrinmaya Sachan, Yinya Huang

NeurIPS 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes DAPS, a dependency-aware premise selector for LLM-based Lean 4 theorem proving that aims to surface newly contributed lemmas buried under popular foundational ones.

Problem

Premise selection for LLM theorem provers in Lean 4: retrieval skews toward frequently used foundational lemmas, so the right premises, especially newly contributed ones, are buried among hundreds of thousands of candidates.

Approach

Moves beyond semantic-similarity neural selectors, which have frequency-skewed retrieval, isolated pairwise ranking, and a third limitation cut off in the abstract; the dependency-aware approach is inferred from the title only and the details are not stated in the available text.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace single-positive InfoNCE with the group-level contrastive objective (ablation)DAPS · Mathlib4-HeldoutRecall@32: 56.00 -> 89.25Improved
add group-level contrastive objective (ablation)DAPS · DAPS-CollectionRecall@32: 15.25 -> 23.53Improved
use DAPS with structural neighborhood encoding instead of LeanHammerpremise selector · Mathlib4-HeldoutRecall@32: 77.84 -> 89.25Improved
DAPS under competition-style query shiftpremise selector · DAPS-CollectionRecall@32: 3.19 -> 23.53Improved
add Blueprint continual training after Mathlib pretrainingDAPS · DAPS-BlueprintRecall@32: 40.42 -> 50.08 vs LeanHammerImproved
use DAPS premises for Goedel-Prover-V2-8BGoedel-Prover-V2-8B · miniF2F-testpass@4: 49.90 -> 54.03Improved
use LeanHammer premises for Goedel-Prover-V2-8BGoedel-Prover-V2-8B · miniF2F-testpass@4: 49.90 -> 48.93Hurt
use DAPS premises for DeepSeek-Prover-V2-7BDeepSeek-Prover-V2-7B · miniF2F-testpass@1: 34.07 -> 38.11Improved
use DAPS premises instead of LeanExplore for Goedel-Prover-V2-8BGoedel-Prover-V2-8B · miniF2F-testpass@1: 40.37 -> 39.96Mixed
supply DAPS top-16 premises to general LLMs for informal provingQwen3-235b, DeepSeek-V3.1, kimi-k2, Gemini-3.6-Flash · IMO-ProofBench Pass@8grade: +0.15 to +3.18 over Qwen3-Embedding-8BImproved
swap formalizer: Gemini-3-Pro to Goedel-Formalizer-V2-32B to manual formalization (ablation)informal proving pipeline · IMO-ProofBench, Gemini-3.6-Flash, Pass@4grade: 46.55 -> 47.50 -> 48.27Improved
add DAPS structural features, rare-premise bucketDAPS vs LeanHammer · Mathlib4-Heldout hit@32 rarest buckethit@32: 39.4 -> 50.1Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy