DeepHyperion: Exploring the Feature Space of Deep Learning-Based Systems through Illumination Search

TL;DR

DeepHyperion employs Illumination Search to systematically explore and visualize the feature space of deep learning systems, aiding in understanding how input features influence system behavior, especially in safety-critical applications.

Contribution

We introduce DeepHyperion, a novel tool that maps the feature space of deep learning systems, enabling interpretability and targeted testing through illumination search techniques.

Findings

Identifies high-risk inputs spread across feature space

Provides interpretable feature maps for DL systems

Facilitates understanding of feature influence on behavior

Abstract

Deep Learning (DL) has been successfully applied to a wide range of application domains, including safety-critical ones. Several DL testing approaches have been recently proposed in the literature but none of them aims to assess how different interpretable features of the generated inputs affect the system's behaviour. In this paper, we resort to Illumination Search to find the highest-performing test cases (i.e., misbehaving and closest to misbehaving), spread across the cells of a map representing the feature space of the system. We introduce a methodology that guides the users of our approach in the tasks of identifying and quantifying the dimensions of the feature space for a given domain. We developed DeepHyperion, a search-based tool for DL systems that illuminates, i.e., explores at large, the feature space, by providing developers with an interpretable feature map where automatically generated inputs are placed along with information about the exposed behaviours.