Reproduce “Kernel Identification Through Transformers” 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
Fergus Simpson, Ian Davies, Vidhi Lalchand, Alessandro Vullo, Nicolas Durrande, Carl Rasmussen
KITT learns to recommend kernel structures for Gaussian-process regression using a transformer trained on synthetic data. For a new dataset, it processes input–output pairs and proposes kernels from a vocabulary; GP hyperparameters are optimized separately. Recommendations take under 0.1 seconds. On Yacht, KITT’s NLPD is −0.25, better than AHGP’s 0.99 but not tree search’s −0.40.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.