Decomposing Representations for Deterministic Uncertainty Estimation
Published in 6th Bayesian Deep Learning workshop at NeurIPS 2021, 2021
Decompose representations for overall better AUROC across multiple benchmarks.
Download here
Publications
Can convolutional ResNets approximately preserve input distances? A frequency analysis perspective
Published in preprint, 2021
A frequency analysis of the bi-Lipschitz constraint in ResNets.
Download here
Feature Space Singularity for Out-of-distribution Detection
Published in SafeAI 2021, 2020
OOD samples concentrate in the feature space.
Download here
Nostalgic Adam: Weighing more of the past gradients when designing the adaptive learning rate
Published in IJCAI 2019, 2019
A simple variant of Adam that solves the theoretical divergence problem and also improves performance
Download here