AI研究偏好模型
AI Research Preference Models
浏览论文内容
中文总结 AI 辅助
该研究针对AI研究智能体的候选方案评估成本过高问题，提出AI研究偏好模型，将其集成到AIRA-dojo智能体后提升了基准任务性能，且达到目标性能的时间更短、预算更少。
中文摘要 AI 辅助
人工智能研究智能体（AIRA）目前能够自主提出、实现和评估自身的机器学习实验，但前沿任务的进展受限于成本：一个候选解决方案可在数分钟内完成编写，而对其进行评估则需耗费数小时至数天的GPU时间。因此，智能体可提出的候选方案数量远超其可承担的运行规模，其进展取决于自身的研究偏好：即如何在众多候选方案间分配固定的执行预算。我们提出AI研究偏好模型（RPMs），用于预测多个候选解决方案中最具执行价值的方案，无需承担全部候选方案的执行成本。我们基于冻结的预训练语言模型构建RPMs（无需进行特定任务训练），分为两种形式：一种是仅用于推理的模型，可对候选方案、代码及先前执行的解决方案进行推理；另一种是智能体模型，在决策前还会运行小规模的试点实验。我们将两种模型集成到AIRA-dojo搜索智能体中，并在AIRS-Bench（一个近期针对人工智能研究智能体的机器学习研究任务基准）上进行评估。两种变体分别将平均归一化得分从0.684提升至0.711和0.729，且使用不到三分之二的执行预算，即可在约15小时内达到无引导智能体24小时的性能。我们的最优RPMs还在两项AIRS-Bench任务上取得了新的 state-of-the-art 结果。
英文摘要
AI research agents (AIRA) can now carry machine learning experiments from proposal through implementation and evaluation. Yet progress on frontier tasks is throttled by the cost of evaluations that can consume days of GPU time. When an agent can propose far more candidates than it can afford to run, progress depends on its research preference: how it allocates a fixed execution budget across many candidates. We introduce AI Research Preference Models (RPMs) that predict which candidate solution is most promising, without paying the cost of running them all. We build RPMs from frozen pretrained language models in two variants: an inference-only model that reasons over candidate plans, code, and previously executed solutions, and an agentic model that additionally runs small-scale pilot experiments. Integrated into the AIRA-dojo research agent and evaluated on the machine learning research benchmark AIRS-Bench, the two variants increase the average normalized score from 0.684 to 0.711 and 0.729, respectively. Both reach the unguided agent's 24-hour performance in roughly 15 hours, using less than two-thirds of its execution budget, and together yield new state-of-the-art results on two AIRS-Bench tasks.
发表机构
- FAIR at Meta
- University of Oxford（牛津大学）
- University College London（伦敦大学学院）
机构由 AI 辅助整理，请以论文原文为准。
