I attended the 2020 American Control Conference, which was supposed to take place in Denver, Colorado, but moved to a 100% online format as a precaution against COVID-19. I normally include a photo of the conference venue with these blog posts, but I don’t think anybody would be interested in seeing my home office setup so I’ll go with no photo this time!
At ACC, I presented a paper with Peter Seiler on the topic of using IQCs for algorithm synthesis. Iterative algorithms like gradient descent can be viewed as robust controllers. In robust control, we have an uncertain dynamical system we would like to control, and we assume the uncertainty is bounded. For example, we could have a mechanical or electrical system consisting of components that are subject to manufacturing tolerances and so we can’t know their exact parameters. Or perhaps their parameters change slowly over time as wear and tear sets in. The goal is to design a feedback controller that works for all systems in this bounded set. This robustness is desirable because if we build 100 identical robots, they won’t actually be identical. Using a robust controller means that we can design a single controller to use in all robots, and it will work even though the robots are not exactly identical. Optimization algorithms are very similar: we want to design an iterative optimization algorithm that works on a variety of different objective functions. The optimization algorithm plays the role of the controller and the objective function plays the role of the uncertain plant. In this paper, we showed how IQC synthesis tools from robust control theory can be used to design robust iterative algorithms.
This year, ACC experimented with a new type of presentation. Rather than the standard 20-minute talk, presenters were given the option to do 3-minute “rapid interactive” talks. I opted for this new format, and since all talks had to be pre-recorded, I’m happy to share with you my 3-minute talk explaining our paper! Here is a link to the video. My short talk was also awarded one of the inaugural “Rapid Interactive People’s Choice Awards”!
Overall, I the conference was very well run and I commend the organizers and staff for getting everything together under such difficult circumstances. I definitely miss the ability to meet with people face-to-face and the spontaneous encounters and discussions that happen at large conferences, but there are also advantages to an online format. There are no logistical/scheduling issues that arise from two talks happening at the same time. And the ability to pause a talk or watch it again later is very nice!