Reproduce “Large Language Models as a Universal Searcher for Open-World Automated Machine Learning” 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
Yuxuan Hu, Jihao Liu, Weikang Shi, Xijie Huang, Linda Wei, Mingjie Zhan, Aojun Zhou, Rui Liu and 1 more
The authors introduce a shared searcher for jointly selecting architectures, training settings, and data mixtures. It represents configurations as task-specific index sequences and trains a language model to rank candidates from optimization trajectories on one vision task. Without task-specific fine-tuning, it transfers to other tasks; text-to-image GenEval reaches 0.91 versus OptFormer’s 0.84, with model sizes listed inconsistently as 1.1B in the table and 1.3B in methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.