[论文] AgentSSL: Can MLE Agents Leverage Unlabeled Data?

论文概要

研究领域: ML
作者: Ethan Lin, Akanksha Sarkar, Ziang Liu, Kristin Branson, Jennifer J. Sun
发布时间: 2026-11-12
arXiv: 2611.17323

中文摘要

将机器学习系统迁移到新领域时，标注数据获取与手工管道工程是两大瓶颈。近期 MLE 智能体在自动化管道设计方面展现出潜力，但研究几乎都在全监督环境下进行，在标注稀缺、未标注数据充裕的真实场景中尚未经过检验。本工作提出智能体化半监督学习（AgentSSL），将 SSL 构建为程序搜索问题，研究智能体在弱监督下如何探索这一空间。我们发现，即使在反馈信号微弱、算力对等的条件下，智能体也能合成代码来组合与改造已有 SSL 技术，在 12 个设置中的 8 个上超越最强的人类工程基线，最高提升 +21.8%。但程序选择被证实为关键瓶颈——若按智能体自身的反馈指标选程序，性能最多损失 47%。为系统研究这一瓶颈，我们发布 ProgramSelectBench，包含 16,977 个智能体生成与人类工程的 SSL 程序的大规模基准，并评估了无监督与主动式程序选择方法。我们还在真实生态学检测任务上评估了 AgentSSL，展示了将智能体发现的策略迁移到受控基准之外的机遇与挑战。

原文摘要

Applying machine learning systems to new domains introduces bottlenecks in both labeled data acquisition and manual pipeline engineering. While recent MLE agents show promise in automating pipeline design, they have been studied almost exclusively in fully supervised settings, leaving them untested in realistic regimes where labeled data is scarce but unlabeled data is abundant. In this work, we introduce Agentic Semi-Supervised Learning (AgentSSL), framing SSL as a program search problem and studying how agents navigate this space under limited supervision. We find that even under weak feedback and matched compute, agents can synthesize code that combines and adapts established SSL techniques, exceeding the strongest human-engineered baselines in 8 of 12 settings by up to +21.8%. However,...

自动采集于 2026-11-14

#论文 #arXiv #ML #小凯