DeepAI
Log In Sign Up

A statistical theory of out-of-distribution detection

02/24/2021
by   Xi Wang, et al.
0

We introduce a principled approach to detecting out-of-distribution (OOD) data by exploiting a connection to data curation. In data curation, we exclude ambiguous or difficult-to-classify input points from the dataset, and these excluded points are by definition OOD. We can therefore obtain the likelihood for OOD points by using a principled generative model of data-curation initially developed to explain the cold-posterior effect in Bayesian neural networks (Aitchison 2020). This model gives higher OOD probabilities when predictive uncertainty is higher and can be trained using maximum-likelihood jointly over the in-distribution and OOD points. This approach gives superior performance to past methods that did not provide a probability for OOD points, and therefore could not be trained using maximum-likelihood.

READ FULL TEXT

page 2

page 5

08/13/2020

A statistical theory of cold posteriors in deep neural networks

To get Bayesian neural networks to perform comparably to standard neural...
08/07/2020

Investigating maximum likelihood based training of infinite mixtures for uncertainty quantification

Uncertainty quantification in neural networks gained a lot of attention ...
12/04/2017

Stochastic Maximum Likelihood Optimization via Hypernetworks

This work explores maximum likelihood optimization of neural networks th...
11/05/2020

Amortized Conditional Normalized Maximum Likelihood

While deep neural networks provide good performance for a range of chall...
09/22/2021

Entropic Issues in Likelihood-Based OOD Detection

Deep generative models trained by maximum likelihood remain very popular...
02/27/2022

Towards Unifying Logical Entailment and Statistical Estimation

This paper gives a generative model of the interpretation of formal logi...
06/10/2021

Data augmentation in Bayesian neural networks and the cold posterior effect

Data augmentation is a highly effective approach for improving performan...