Reproduce “Conditional Flow Models Convert Epistemic Uncertainty into Apparent Aleatoric Variance” 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
Navodita Sharma, Abhishek Shetty, Max Simchowitz, Pulkit Agrawal
The authors study how conditional flow models represent uncertainty in nearly deterministic prediction tasks. They propose that smoothing across nearby inputs can leave dependencies unresolved, producing extra sample variance that is indistinguishable from genuine data noise. In experiments, learned variance falls as smoothing decreases; predictive sets are approximately calibrated during underfitting, but coverage is empirical, not guaranteed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.