Dissertation Proposal by Imam Nur Bani Yusuf | Advancing Domain-specific Code Generation using Transformer-based Deep Learning

Software has become an integral part of our daily lives as digitalization becomes increasingly pervasive. The architects of this digital landscape are developers who write lines of code to define the logic of these software systems. However, developers face challenges in writing code regardless of their experience level. These challenges often stem from the lack of familiarity with the elements within the coding. To tackle this challenge, the dissertation aims to develop automated code generation tools using Transformer-based deep learning to assist developers in writing domain-specific codes.

This dissertation comprises three distinct studies. The first introduces RecipeGen, an automated tool for creating trigger-action programs from natural language descriptions. The second proposes ArduinoProg, a framework designed to automate recommendations for I/O hardware configurations and API usage patterns. Evaluation of both RecipeGen and ArduinoProg shows their effectiveness in aiding code development. The third study investigates the use of language models for generating semantic patches in C code within the Linux kernel. Evaluation reveals potential areas for improvement in deep learning models.

Imam Nur Bani Yusuf is a Ph.D. candidate at SMU SCIS, supervised by Associate Professor Lingxiao Jiang and co-advised by Professor David Lo. His research is centered around Language Models for Software Engineering (LM4SE). His interests span from developing novel language models tailored to address specific problem domains, to enhancing the capabilities of language models through techniques such as fine-tuning and prompting. More info available: https://imamnurby.github.io/.