SELA
Object
SELA (Tree-Search Enhanced LLM Agents) is an innovative system that enhances Automated Machine Learning (AutoML) by integrating Monte Carlo Tree Search (MCTS) with LLM-based agents to overcome the limitations of traditional AutoML approaches.
Features
- Monte Carlo Tree Search (MCTS) optimization framework for exploring machine learning solution spaces
- Tree-structured representation of pipeline configurations for intelligent exploration
- Iterative refinement of strategies based on experimental feedback
- Support for multiple datasets with configurable metrics and parameters
- Flexible rollout system with adjustable depth and timeout settings
- Ability to resume interrupted experiments through tree loading
- Comprehensive ablation study capabilities with random search modes
- Integration with various machine learning datasets and tasks
Outcome
It demonstrates exceptional performance across twenty machine learning datasets, achieving win rates ranging from sixty-five percent to eighty percent compared to baseline AutoML methodologies. This advancement facilitates a more effective exploration of the machine learning solution space while producing higher-quality and more diverse code solutions than conventional approaches, ultimately empowering artificial intelligence to develop optimal AI systems autonomously.