On April 6th at 2:30 PM, Berkeley Research Computing (BRC) reached a new milestone: the 1000th user account created in the high performance computing (HPC) environment. The lucky recipient of the 1000th account was Ted Xiao, an M.S. student working with Prof. Claire Tomlin in the Electrical Engineering and Computer Science department. In addition to working on machine learning and control systems, he is actively involved with Machine Learning at Berkeley, a student-run organization he founded last year.
Ted’s account was created as a member of Professor Tomlin’s faculty computing allowance (FCA) group. FCAs provide 300,000 “service units” on the Savio cluster annually to any qualified faculty member or principal investigator. Faculty members with an FCA can give students, postdocs, or other researchers access to BRC’s Savio cluster by authorizing creation of a user account associated with their FCA.
Using Savio will be a significant step forward in Ted’s research: “I'm looking forward to using Savio for a Deep Reinforcement Learning project. The project focuses on providing a ‘guiding hand’ to model-free reinforcement learning, through the use of auxiliary image-based tasks. Specifically, image observations will be predicted using a Generative Adversarial Network, a fairly recent model that has shown great promise in producing more realistic, concise images. I'll be using Savio's GPU [nodes] to train the neural network models, which will be a great boost in efficiency. What would take days or weeks on my personal computer can be trained in hours on Savio.”
Since the launch of BRC in summer 2014, the program has added more than a user per day, on average. New FCA requests have accelerated in the last few months, with 3-5 requests coming in per week. HPC environments are only one component of BRC: several hundred researchers are supported by the program’s cloud consulting services, and the recently-launched Analytic Environments on Demand (AEoD) virtual research desktop service is nearing a hundred users.