Sponsored by:
NSF logo
Latest news:
Important Dates:
Contact Information:
Oakland University

2024 Projects

Advisor: Dr. Siadat
Advisor: Dr. Wardat
This project explores the use of transformer models for automating bug detection and diagnosis in deep neural networks (DNNs). DNNs are widely used in critical applications like autonomous vehicles and medical diagnosis, where accuracy and reliability are crucial. However, debugging these models is challenging and resource-intensive. Traditional trial-and-error methods are time-consuming and inefficient. To address this, we leveraged transformer models such as PLBART, CodeBERT, and GraphCodeBERT for feature extraction from both faulty and correct models. These features were then used to train deep learning models to classify whether a model contains bugs and to identify the type of bug. Our goal is to create a tool that assists developers in quickly identifying and fixing issues in their AI models, thereby enhancing development efficiency and improving model performance.