David OBrien

Research Interests

Software Engineering

Machine Learning

Program Analysis

Big Data Mining

Publications

• David OBrien, Sumon Biswas, Sayem Imtiaz, Rabe Abdalkareem, Emad Shihab, Hridesh Rajan. Are Prompt Engineering and TODO Comments Friends or Foes? An Evaluation on GitHub Copilot. In 46th International Conference on Software Engineering (ICSE), 2024.
• David OBrien, Robert Dyer, Tien Nguyen, Hridesh Rajan. Data-Driven Evidence-Based Syntactic Sugar Design. In 46th International Conference on Software Engineering (ICSE), 2024.
• David OBrien, Sumon Biswas, Sayem Imtiaz, Rabe Abdalkareem, Emad Shihab, Hridesh Rajan. 23 Shades of Self-Admitted Technical Debt: An Empirical Study on Machine Learning Software. In 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE), 2022.

Previous Experiences

• Laboratory for Software Design (Summer 2020 - Present)

  • Advocate for and researched self-admitted technical debts in open-source machine learning repositories. Research has supplied empirical evidence for technical debt types proposed by prior works, and discovers new ML-specific technical debts such as dependency, interpretability, and knowledge debts. Work has been described as "pushing forward state of the art knowledge of machine learning software".
  • Implement new functionality to the Boa Language and Infrastructure, including the implementation of scalable frequent subgraph mining algorithms with Hadoop's MapReduce framework. Contributions enable researchers to mine subgraphs from large-scale datasets.

• Apple - ML Engineer Intern (Summer 2023)

  • Implemented a regression testing computer vision framework to analyze multi-versioned UI screenshots and quantify their differences. Evaluations shows our technique of deep-learning-enabled difference identification is 10x faster and 50% more correct than manual evaluation.
  • Experimented with multiple pre-trained architectures and fine-tuned convolutional neural networks to determine best solution for business problem.
  • Manually curated a dataset of multiple screenshots representing regression-inducing changes and non-regression-inducing changes.

• Collins Aerospace - AI Engineer Intern (Summer 2022)

  • Maintain multi-year text-extracting software originally valued at 9.8 million dollars. Contributions include identifying and resolving defects, removing obsolete code, and leading discussions on worthwhile coding standards to adopt to reduce developer maintenance waste and promote easily maintained software.
  • Apply supervised learning processing techniques on unseen data to provide clients with organized data from AWS.
  • Experimented with reinforcement learning to guide a simulated car through a course which consistently exceeded score expectations to motivate autonomous systems research.
  • Cleaning, organizing, and evaluating a dataset consisting of multiple real-time sensors used by future engineers.