[PAST EVENT] Computer Science Alumni Colloquium Series: Bruce Childers '91
Access & Features
- Open to the public
Achieving Scientific Reproducibility for Experimental Computer Science
Bruce Childers (BS '91)
University of Pittsburgh
Experimental computer science relies on a remarkably rich and diverse set of software and data to implement and evaluate new ideas on many measures, such as accuracy, performance, energy, and resilience. Unfortunately, ever-increasing competitiveness and pressure to produce more and more results pose an impediment to accountability, which is a critical part of the scientific method. Further, experiments, along with their constituent software and data, may not be disseminated in a way to achieve the repeatability and/or reproducibility needed for trust, fairness and building on and comparing with previous research. In this talk, I will advocate for a solution to this "reproducibility crisis" that requires culture change and technology advancement. Experimental computer science needs to dramatically realign its perception of the value and reward of producing and openly sharing high-quality software and data. Artifact Evaluation is one way to incentivize the community to bring about this change. Technological advances are also needed to seamlessly connect artifacts to all steps of the research lifecycle for easier and more transparent sharing. I will describe and demonstrate a novel platform, OCCAM, that enables this type of sharing by directly linking experiments to the scholarly record as interactive, modifiable and executable content.
Bio: Bruce Childers (W&M BS ?91, UVA PhD ?00) is a Professor of Computer Science at the University of Pittsburgh. His research spans the software-hardware boundary for improved energy consumption, performance and reliability. He has developed techniques at both the software layer (dynamic binary translation, compiler optimization, debugging and software testing) and the hardware layer (speed scaling, reliable cache, storage class memory, and GPUs). It was during research on computer architecture that he grew frustrated with re-implementing published work, which was often significantly underspecified, for comparison. This frustration turned into positive action to develop approaches for reproducible science and to demonstrate why accountable research is important. Childers is the past steering committee and program chair of the ACM SIGPLAN/SIGBED Conference on Languages, Compilers, and Tools for Embedded Systems, participates on an ACM task force for scientific reproducibility, led a pilot to connect active curation platforms with the ACM Digital Library, and one of the contributors to proposing Pitt?s new School for Computing and Information, which started July 2017.