Grant: Research Tools and Instruments

In an era marked by rapid advancements in large pre-trained models (PTMs) for automating software engineering (SE) tasks, a critical necessity arises for a fully configurable, scalable computing platform to evaluate and optimize the capabilities of PTMs in SE domains in the context of research collaborations with Canadian industry. The specialized PTM computing platform outlined by this proposal bridges this technological gap and nurtures innovation within Canadian SE industry, while fostering a rich, hands-on learning environment for the highly qualified personnel (HQP) at Queen's University. The proposed platform is equipped with high-performance GPU servers, aiding in prompt engineering, fine-tuning, and empirical evaluation of open-source large PTMs for SE tasks. The equipment will serve as the backbone for three key research activities: advancing novel prompt engineering and fine-tuning methodologies, enabling controlled, secure empirical evaluation and selection of PTMs, and ensuring effective deployment and monitoring of optimized PTMs. These activities are unique and fundamental for addressing the challenges faced by Canadian companies in optimizing PTMs to their specific contexts, which is often resource-intensive and costly. The proposed platform will significantly impact the training and practical exposure of around 30 HQP annually, across two labs directly and several other SE and AI research labs indirectly within the School of Computing. Through collaborative workshops, a graduate-level course, and real-world SE challenges, the platform will provide hands-on experience in managing (via direct access to the infrastructure of this computing platform), deploying, and optimizing PTMs, thereby furnishing HQP with marketable skills essential in the intertwined realms of SE and AI.