KompeteAI: New AutoML Framework with Multi-Agent System
A new research paper introduces KompeteAI, an AutoML framework designed to overcome limitations in existing Large Language Model (LLM)-based systems. The system addresses two key issues: constrained exploration strategies and execution bottlenecks. For exploration, KompeteAI uses a merging stage to combine top candidates from Monte Carlo Tree Search (MCTS), unlike previous methods that treat ideas in isolation. It also integrates Retrieval-Augmented Generation (RAG) to source ideas from Kaggle notebooks and arXiv papers. To tackle execution bottlenecks, the framework employs accelerated autonomous multi-agent pipelines for end-to-end machine learning problem solving. The paper is available on arXiv under ID 2508.10177.
Key facts
- KompeteAI is an AutoML framework for machine learning problems.
- It uses a multi-agent system for end-to-end pipeline generation.
- Addresses exploration limitations of one-shot methods and MCTS.
- Introduces a merging stage to compose top candidates.
- Integrates RAG from Kaggle and arXiv for hypothesis expansion.
- Aims to reduce execution bottlenecks from lengthy code validation.
- Paper published on arXiv with ID 2508.10177.
- Focuses on LLM-based AutoML systems.
Entities
Institutions
- arXiv
- Kaggle