I attended the 2018 American Control Conference in Milwaukee, WI. This was a unique opportunity, being a local conference (only a short bus ride away from Madison!) so I took the opportunity to bring the whole research group (photo on the right).
At the conference, my student Saman Cyrus presented our paper (also co-authored by Bin Hu and Bryan Van Scoy) titled “A robust accelerated optimization algorithm for strongly convex functions”. In this work, we proposed an accelerated algorithm with a single tuning parameter that controls the trade-off between convergence rate and robustness to multiplicative gradient noise. We call it the “Robust Momentum Method” (RMM). Existing algorithms tend to be “slow and robust” like Gradient Descent or “fast and fragile” like the Heavy-Ball method or Nesterov’s method. With RMM, the single tuning parameter directly controls this trade-off, leading to intermediate options that are both fast and robust. This was Saman’s first conference presentation and he gave an excellent talk! His slides are available here.
I enjoy attending seminars outside my field, and there were two excellent plenaries at ACC this year that fit the bill. Robert Wood from Harvard spoke about the engineering challenge of building micro-robots that mimic flying insects. It’s a truly multi-disciplinary endeavor combining micro-fabrication, fluid mechanics, structural mechanics, material science, electronics, control engineering, and origami! The second talk that really expanded my horizons was by Emory Brown from Harvard Medical School and MIT. Emory is a professor of Medical Engineering, Computational Neuroscience, Health Sciences and Technology, and Anesthesia. He is also a practicing anesthesiologist and has a PhD in statistics. The talk was about controlling one of the most complicated things in the world: the human brain.