Input complexity and out-of-distribution detection with likelihood-based generative models

TL;DR

This paper investigates how input complexity affects likelihood-based generative models' ability to detect out-of-distribution inputs and proposes a new likelihood-ratio score that improves detection performance across various datasets and models.

Contribution

It identifies input complexity as a key factor influencing likelihoods and introduces a parameter-free likelihood-ratio score for better OOD detection.

Findings

The proposed score performs comparably or better than existing methods.

Input complexity significantly impacts likelihood-based OOD detection.

The method is effective across diverse datasets and model configurations.

Abstract

Likelihood-based generative models are a promising resource to detect out-of-distribution (OOD) inputs which could compromise the robustness or reliability of a machine learning system. However, likelihoods derived from such models have been shown to be problematic for detecting certain types of inputs that significantly differ from training data. In this paper, we pose that this problem is due to the excessive influence that input complexity has in generative models' likelihoods. We report a set of experiments supporting this hypothesis, and use an estimate of input complexity to derive an efficient and parameter-free OOD score, which can be seen as a likelihood-ratio, akin to Bayesian model comparison. We find such score to perform comparably to, or even better than, existing OOD detection approaches under a wide range of data sets, models, model sizes, and complexity estimates.