Reproduce “DiffLTM: Continuous Diffusions Large Tabular Models for Conditional Generation” 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
Alankrita Bhatt, Mukur Gupta, Andrea Montanari, Pulkit Tandon
DiffLTM generates mixed-type tabular data conditioned on exact values, numerical ranges, or linear constraints. It embeds categories and numerical values in a continuous representation and uses classifier-free guidance for conditional generation. On ACSLabour, categorical cross-entropy with logits scores 94.87 versus 81.40 for mean-squared error on equal-weighted subgroup fidelity. The Arizona counterfactual remains unverified because its ground truth is unknown.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.