SELA is an open-source agent that autonomously designs AI models, harnessing the power of Monte Carlo Tree Search (MCTS) to achieve state-of-the-art performance across 20 machine learning datasets. Unlike traditional AutoML frameworks that use fixed pipelines, SELA uses a dynamic, tree-based approach to represent and optimize pipeline configurations, making it adaptive and capable of evolving through iterative design improvements.

SELA evaluates and learns from previous designs to improve performance, leveraging experimental feedback to explore solution spaces intelligently. This allows SELA to create diverse, high-performing pipelines in scenarios where conventional AutoML agents struggle, particularly in generating optimal code diversity. By demonstrating a win rate of 65%-80% against baselines, SELA sets a new benchmark for agent-based AutoML in finding optimal machine learning pathways.

Key Features:

AI-Driven AI Design: Uses MCTS to design and refine AI models autonomously, improving through past experiments.
Dynamic AutoML Pipelines: Constructs flexible pipelines that adapt based on agent feedback, enhancing solution diversity.
High Win Rate: Outperforms traditional AutoML methods with a win rate of up to 80% across datasets.
Open-Source and Accessible: Fully open-source with accessible code on GitHub, fostering collaborative AI development.
Research Ready: Supported by detailed documentation and a comprehensive research paper, making it ideal for advanced machine learning research.

SELA offers a pioneering framework for AI-driven AI design, ideal for researchers and developers aiming to streamline machine learning optimization with intelligent, automated systems.

Agents

AI Tools

SELA