Towards neural networks that provably know when they don't know

09/26/2019
by   Alexander Meinke, et al.
0

It has recently been shown that ReLU networks produce arbitrarily over-confident predictions far away from the training data. Thus, ReLU networks do not know when they don't know. However, this is a highly important property in safety critical applications. In the context of out-of-distribution detection (OOD) there have been a number of proposals to mitigate this problem but none of them are able to make any mathematical guarantees. In this paper we propose a new approach to OOD which overcomes both problems. Our approach can be used with ReLU networks and provides provably low confidence predictions far away from the training data as well as the first certificates for low confidence predictions in a neighborhood of an out-distribution point. In the experiments we show that state-of-the-art methods fail in this worst-case setting whereas our model can guarantee its performance while retaining state-of-the-art OOD performance.

READ FULL TEXT

page 2

page 6

research
12/13/2018

Why ReLU networks yield high-confidence predictions far away from the training data and how to mitigate the problem

Classifiers used in the wild, in particular for safety-critical systems,...
research
06/08/2021

Provably Robust Detection of Out-of-distribution Data (almost) for free

When applying machine learning in safety-critical systems, a reliable as...
research
11/17/2021

Do Not Trust Prediction Scores for Membership Inference Attacks

Membership inference attacks (MIAs) aim to determine whether a specific ...
research
10/06/2020

Fixing Asymptotic Uncertainty of Bayesian Neural Networks with Infinite ReLU Features

Approximate Bayesian methods can mitigate overconfidence in ReLU network...
research
11/29/2022

Birds of a Feather Trust Together: Knowing When to Trust a Classifier via Adaptive Neighborhood Aggregation

How do we know when the predictions made by a classifier can be trusted?...
research
05/29/2018

Representational Power of ReLU Networks and Polynomial Kernels: Beyond Worst-Case Analysis

There has been a large amount of interest, both in the past and particul...
research
11/10/2022

Quorum Systems in Permissionless Network

Fail-prone systems, and their quorum systems, are useful tools for the d...

Please sign up or login with your details

Forgot password? Click here to reset