Reproduce “Temporal Causal Prior-Data Fitted Networks for Panel Data with Learned Reliability Signals” 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

Shravan Talupula, Saurabh Sharma

TCPFN is a pretrained model for estimating treatment effects and ranking causal links across datasets without fitting each one. It learns from synthetic causal time series, represents unit-time observations as tokens, and keeps prediction horizons separate. On Tennessee Eastman, its AUROC is 0.959 versus PCMCI’s 1.000, but oracle-threshold F1 is 0.613 versus 1.000, supporting screening rather than formal causal identification.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.