AutoReproduce: Automatic AI Experiment Reproduction with Paper Lineage

arXiv cs.AI / 4/27/2026

📰 NewsIdeas & Deep AnalysisTools & Practical UsageModels & Research

共有:

Key Points

The paper introduces “paper lineage,” an approach that mines implicit knowledge from a paper’s cited literature to make research reproduction less dependent on domain expertise.
It presents AutoReproduce, a multi-agent framework that autonomously reproduces experimental code end-to-end, aiming for complete workflow coverage.
To improve executability, AutoReproduce uses a sampling-based unit testing strategy for fast validation during reproduction.
The authors propose “AutoReproduceBench” (referred to as ourbench), a benchmark with verified implementations and metrics to evaluate both reproduction fidelity and execution fidelity.
Experiments on PaperBench and AutoReproduceBench show AutoReproduce outperforms prior baselines across all metrics, with notable gains in both reproduction fidelity and final execution performance.

Abstract

Efficient reproduction of research papers is pivotal to accelerating scientific progress. However, the increasing complexity of proposed methods often renders reproduction a labor-intensive endeavor, necessitating profound domain expertise. To address this, we introduce the paper lineage, which systematically mines implicit knowledge from the cited literature. This algorithm serves as the backbone of our proposed \ours, a multi-agent framework designed to autonomously reproduce experimental code in a complete, end-to-end manner. To ensure code executability, \ours incorporates a sampling-based unit testing strategy for rapid validation. To assess reproduction capabilities, we introduce \ourbench, a benchmark featuring verified implementations, alongside comprehensive metrics for evaluating both reproduction and execution fidelity. Extensive evaluations on PaperBench and \ourbench demonstrate that \ours consistently surpasses existing baselines across all metrics. Notably, it yields substantial improvements in reproduction fidelity and final execution performance. The code is available at https://github.com/AI9Stars/AutoReproduce.