Benchmarking Generative AI Models for Deep Learning Test Input Generation
This work addresses the problem of efficiently generating high-quality test inputs for deep learning systems, which is incremental as it benchmarks and combines existing methods rather than introducing new ones.
The paper benchmarks Generative AI models for generating test inputs to evaluate deep learning image classifiers, finding that simpler models like VAEs suffice for simple datasets like MNIST, while more complex models like Diffusion Models perform better on feature-rich datasets like ImageNet, generating more valid inputs that cause misclassifications.
Test Input Generators (TIGs) are crucial to assess the ability of Deep Learning (DL) image classifiers to provide correct predictions for inputs beyond their training and test sets. Recent advancements in Generative AI (GenAI) models have made them a powerful tool for creating and manipulating synthetic images, although these advancements also imply increased complexity and resource demands for training. In this work, we benchmark and combine different GenAI models with TIGs, assessing their effectiveness, efficiency, and quality of the generated test images, in terms of domain validity and label preservation. We conduct an empirical study involving three different GenAI architectures (VAEs, GANs, Diffusion Models), five classification tasks of increasing complexity, and 364 human evaluations. Our results show that simpler architectures, such as VAEs, are sufficient for less complex datasets like MNIST. However, when dealing with feature-rich datasets, such as ImageNet, more sophisticated architectures like Diffusion Models achieve superior performance by generating a higher number of valid, misclassification-inducing inputs.