Article Published In Vol.9 (Jan-Feb 2021)

Lightweight Adversarial Software Testing via CNN and Particle Swarm

Author : Venkata Sivakumar Musam and Karthick.M

Convolutional neural networks (CNNs), in particular, are machine-learning models that are increasingly being utilised in modern software development to forecast problems and guide testing procedures. The fundamental issue that this paper addresses is how to conduct effective adversarial software testing in environments where there are hard resource constraints. Current methods whether deep neural networks or exhaustive search—yield high-quality adversarial inputs but require a large amount of computation, memory, and latency. This work presents a lightweight adversarial testing methodology that blends CNNs and PSO to ensure efficient and goal-oriented detection of software vulnerabilities. The innovation here is in its lightweight and efficient nature with a unique combination of Convolutional Neural Networks and Particle Swarm Optimization to conduct adversarial software testing. This integration facilitates efficient defect detection with substantially less computational complexity thus, it is appropriate for real-time and resource-limited scenarios. Experimental outcomes show that the CNN-PSO model is more efficient and effective compared to conventional adversarial testing techniques. On benchmark datasets including PROMISE and NASA MDP, the model recorded an average accuracy of 93.7%, F1 score of 0.91, and adversarial attack success rate of 87.5% while improving computational time reduction to 40% compared to conventional deep learning-based methods. Current adversarial software testing techniques tend to depend on intricate deep learning models or exhaustive search algorithms, which, although efficient, require high computational resources and time. The resultant hybrid solution achieves an interesting balance of precision and performance, demonstrating superiority over existing methods in terms of speed, efficiency, and portability to low-power or embedded platforms at no expense in defect detection efficacy.

Keywords: Software Testing, Convolutional Neural Network, Particle Swarm Optimization