Reproduce “AIDE: AI-Driven Exploration in the Space of Code” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhengyao Jiang, Dominik Schmidt, Dhruv Srikanth, Dixing Xu, Ian Kaplan, Deniss Jacenko, Yuxiang Wu
AIDE has an LLM generate executable machine-learning code, evaluate each attempt, and organize scored candidates in a tree for iterative debugging and improvement. It summarizes relevant history rather than relying on one running transcript. With o1-preview on low-complexity MLE-Bench Lite, any-medal results rose from 7.6% ± 2.6% to 36.4% ± 7.9%. This compares the full scaffold, not tree search alone.
Key insights
- Framing ML engineering as code optimization rather than hyperparameter tuning leads to more flexible and powerful solutions
- Hierarchical organization of solution space enables more efficient exploration than linear approaches
- Combining structured search with LLM capabilities creates synergistic improvements in solution quality
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.