Gunner Levi Howe

Label-free invariance priors can outperform traditional supervised methods, achieving 15/15 successful generalizations with a remarkable 2.7x speedup.

Importance weighting can't fix selection bias, but a Heckman correction can restore predictive accuracy in the face of unobservable data selection.

Achieving a 250x speedup in topology estimation for neural fields could revolutionize how we approach implicit neural representations in high-dimensional spaces.