Lin Tan receives an ACM Distinguished Paper award at ASE 2020
10-02-2020
Professor Lin Tan and fellow researchers win ACM SIGSOFT Distinguished Paper Award at ASE 2020, the 35thIEEE/ACM International Conference on Automated Software Engineering. Tan’s paper titled Problems and Opportunities in Training Deep Learning Software Systems: An Analysis of Variance is one of the first studies to examine variance in deep-learning models and the extent to which computer science researchers and computing professionals are aware of this variance.
Artificial intelligence tools, like deep learning (DL) frameworks, offer building blocks for designing, training and validating raw data, through a high-level programming interface. These tools are useful for researchers and practitioners interested in the outcome data. Complications may arise because deep-learning training algorithms and implementations are not deterministic. This means variances are inevitable and produce different models with different levels of accuracy and training time.
In their recent paper, Professor Lin Tan and fellow researchers proved this variance from the implementations to be unknown to the vast majority of researchers and practitioners. This paper is among the first to study and raise awareness on the variance of DL systems. It also directs software engineering researchers to challenging tasks such as creating deterministic DL implementations to facilitate debugging and improving the reproducibility of DL software and results.
“We hope our award-winning paper will continue raising awareness of deep learning variance, which poses opportunities as well as challenges for researchers and practitioners.” Tan adds, “I am excited to receive support and feedback from NSF, Facebook, and Microsoft to continue these exciting projects on studying and testing the variance and uncertainty of deep learning implementations. We are currently building tools to help researchers and practitioners test the variance of their deep learning systems and reproduce their research and results.”
The distinguished paper is joint work between Professor Tan and PhD students advised by her: Shangshu Qian, Jiannan Wang, and Jonathan Rosenthal at Purdue University, Department of Computer Science, Hung Viet Pham and Thibaud Lutellier, at the University of Waterloo, Professor Yaoliang Yu at the University of Waterloo, and Nachiappan Nagappan from Microsoft Research.
This paper is supported by a Facebook 2020 Probability and Programming research award and NSF grant 2006688.
Citation
Hung Viet Pham, Shangshu Qian, Jiannan Wang, Thibaud Lutellier, Jonathan Rosenthal, Lin Tan, Yaoliang Yu, and Nachiappan Nagappan. Problems and opportunities in training deep-learning software systems: an analysis of variance. ASE 2020 Research Papers.
Writer: Emily Kinsell, emily@purdue.edu,
Source: Professor Lin Tan, lintan@purdue.edu