PACE: Unleashing the Power of Code Embeddings to Boost AutoML Agents
Gangyi Zhao, Hebin Liang, Hongyao Tang, Yi Ma, Jinyi Liu, Zhaocheng Du, Yan Zheng, Chenjun Xiao, Jianye Hao
Abstract
Large Language Model (LLM)-driven AutoML agents have shown strong capabilities in constructing end-to-end machine learning pipelines. However, their effectiveness is limited by costly execution-based feedback, which can make the search for high-quality solutions inefficient under restricted computational budgets. We propose PACE (Pre-execution Admission via Code Embeddings), an online-adaptive admission control framework that improves budgeted sample efficiency by estimating candidate utility prior to execution from within-run execution history, without training a separate offline predictor. The core idea of PACE is to leverage latent structure in the solution space as a within-run admission signal. It projects candidate solutions into multi-view semantic embeddings, dynamically organizes executed candidates into clusters, and estimates new candidates by their proximity to historical elite regions. Moreover, PACE aggregates multi-view embeddings via an adaptive reweighting strategy that prioritizes views with higher discriminative power. This enables PACE to bias the agent toward high-potential regions under a limited computational budget while retaining exploration when local structure is weak. We demonstrate that PACE operates as a plug-and-play admission layer for AutoML agents, serving as either an execution gate or a search prior. In the tested settings, it improves the density of elite solutions found within fixed budgets without modifying the underlying agent architecture. Code, configurations, and prompt templates are publicly available at https://github.com/fendss/PACE.
Ask Lune
Know where this paper stands in minutes, not an afternoon.
Lune can read the top-tier papers that cite it and the ones closest to it, and tells you what it claims, what came after and where it falls short, each point cited to the paper it comes from. No skimming, no AI slop.
Related papers
- Multi-agent Architecture Search via Agentic SupernetGuibin Zhang, Luyang Niu, Junfeng Fang, Kun Wang et al.ICML 2025
- ML-Agent: Reinforcing LLM Agents for Autonomous Machine Learning EngineeringZexi Liu, Jingyi Chai, Xinyu Zhu, shuo tang et al.ICML 2026
- SPaCe: Unlocking Sample-Efficient Large Language Models Training With Self-Pace Curriculum LearningVan Dai Do, Manh Nguyen, Svetha Venkatesh, Hung LeACL Findings 2026
- PRompt Optimization in Multi-Step Tasks (PROMST): Integrating Human Feedback and Heuristic-based SamplingYongchao Chen, Jacob Arkin, Yilun Hao, Yang Zhang et al.EMNLP 2024 · 6 citations
- Do We Always Need Query-Level Workflows? Rethinking Agentic Workflow Generation for Multi-Agent SystemsZixu Wang, Bingbing Xu, Yige Yuan, Huawei Shen et al.ACL Findings 2026
