Goodbye Accounting, Hello High-performance Computing

Janaan Lake is living proof that it’s never too late to pursue a career in computing. After working 17 years as a Certified Public Accountant, she decided the time was right to pursue a computer science degree and enrolled at the University of Utah. “Although changing careers in midlife has been more challenging than I anticipated, it has also been more rewarding.” 

As a non-traditional student, Janaan had the experience to notice and eagerly take advantage of unique opportunities offered to students. She joined the University of Utah’s “SupercompUtes” team for the Student Cluster Competition at the Supercomputing Conference (SC) in 2016 and 2017. One of the competition’s tasks was a reproducibility challenge in which teams replicated the methodology of an SC paper from the previous year and compared results. Team reports from the reproducibility challenge were published in a special issue of Parallel Computing each year, and Janaan was the lead author for the SupercompUtes’ 2017 report. Janaan also had the opportunity to attend presentations and network at SC16 and SC17, expanding her research interests. 

After the competition, Prof. Mary Hall, one of the University of Utah School of Computing faculty who advised the Student Cluster Competition team, invited Janaan to join her Compiler Technology to Optimize Performance (CTOP) research group. Janaan first worked on an autotuning project in collaboration with the U.S. Department of Energy (DOE). Her autotuning helped reduce the search space for parameter values used in SuperLU, a DOE application for directly solving large, sparse, nonsymmetric systems of linear equations. Janaan ran experiments for the project on the Cori supercomputer at the National Energy Research Science Scientific Computing Center (NERSC). This experience motivated her to complete a bachelor’s thesis advised by Prof. Hall.

After sifting through several possible bachelor’s thesis topics, Janaan and Prof. Hall settled on optimized CPU code for Deep Neural Networks (DNNs). Janaan started the project in summer 2019 with some background research on LATTE, a programming language designed specifically for DNNs, and SWIRL, a high-performance compiler for producing optimized C++ code from LATTE. She spent the next semester extending the LATTE/SWIRL workflow to support batch normalization, a technique for decreasing training time and increasing robustness of DNNs. She used a university supercomputer to perform experiments comparing her LATTE/SWIRL batch normalization operator against well-known general-purpose frameworks such as TensorFlow, PyTorch, and MXNet, and she found performance gains for a variety of DNN sizes. The resulting research paper was accepted to the 33rd Workshop on Languages and Compilers for Parallel Computing (LCPC 2020).

Janaan found support throughout her research project from her advisor, Prof. Hall, and a graduate student mentor with whom she met weekly to brainstorm ideas, discuss progress, and solve issues. When asked about her favorite aspect of research, Janaan points to the creativity and open-endedness of research. “In a classroom setting, there are often ‘right’ and ‘wrong’ answers. But in a research setting, that dichotomy is a lot more fluid. Rather than right vs. wrong, the questions seem to be ‘why’ or ‘why not?’ My favorite aspect of research is the challenge of trying to answer those questions. It definitely gets me out of my comfort zone, but that can also be quite rewarding.”

Janaan’s undergraduate research experience spurred her on to pursue a master’s degree in computer science this year, also at the University of Utah. Outside of research and coursework, she enjoys spending time with her husband and three sons, especially when the conditions are right to get outside for rock climbing. To aspiring undergraduate researchers, she emphasizes the importance of building relationships with professors. “Show professors who are conducting research in your area of interest that you are willing to work hard and are self-motivated. Take their classes. Ask them about their research and inquire if there are any projects that you could assist with. Research takes a lot of self-discipline, time management skills and persistence. Develop those traits, and they will serve you well.”

 — Written and edited by Ian Ludden

Leave a comment

Your email address will not be published. Required fields are marked *