DeepRNG: Towards Deep Reinforcement Learning-Assisted Generative Testing of Software

TL;DR

This paper introduces DeepRNG, a framework that enhances software testing by integrating deep reinforcement learning with the random number generator, leading to more effective testing of complex software systems.

Contribution

DeepRNG is the first approach to augment RNG with RL for software testing, demonstrating significant improvements on a large-scale software library.

Findings

Statistically significant testing improvements with DeepRNG

Effective augmentation of RNG with RL for complex software

Open-source implementation of the DeepRNG framework

Abstract

Although machine learning (ML) has been successful in automating various software engineering needs, software testing still remains a highly challenging topic. In this paper, we aim to improve the generative testing of software by directly augmenting the random number generator (RNG) with a deep reinforcement learning (RL) agent using an efficient, automatically extractable state representation of the software under test. Using the Cosmos SDK as the testbed, we show that the proposed DeepRNG framework provides a statistically significant improvement to the testing of the highly complex software library with over 350,000 lines of code. The source code of the DeepRNG framework is publicly available online.