After working on several advanced electron accelerators around the world, Bay Area native Simon Leemann came back home in 2017. He now serves as Deputy for Accelerator Operations and Development at Berkeley Lab’s Advanced Light Source, where ATAP provides the accelerator-physics team.
Most recently his interests have turned to applying machine learning to stabilizing beams at the ALS, then to incorporating these techniques in to the design process for future accelerators.
Join us for three questions drawn from a wide-ranging discussion of how a colleague with enthusiasm for a new idea can send you in a promising direction, how artificial intelligence will transform work, and the value of cultural and intellectual diversity…
“It’s always very enriching to have someone come into the group from outside and make you realize, ‘There’s a very different way of looking at this problem.'”
Machine learning is one of the hot new tools in accelerators and many other areas. How did you get involved?
It’s funny how I pretty much stumbled into it. In our group, Hiroshi Nishimura had been telling us about it, and showed us examples of how it could be used to predict things. After a while, we realized how this could be used to anticipate problems and correct them before they occur.
Hiroshi and I had been working on the same topic for many years (unbeknownst to him). He was the original author of the Tracy tracking code, which I had been using since the early 2000s to do tracking studies. We also had a mutual friend in Etienne Forest, who had also been in AFRD and is now with KEK in Japan.
So while we had never actually met or talked, our work had been closely related all along. When I came to Berkeley, we had a lot of catching up to do! It was natural for us to hit it off here and start collaborating on something new and different. He was so enthusiastic about the idea that I figured, “why not give it a try?”
To be perfectly honest with you, I was a bit skeptical at first, but it turns out that machine learning isn’t really some kind of magical black box — you can learn how the box works, and understand why in some cases it works and in other cases it doesn’t. It’s not the way we’re used to attacking a problem — by taking a physics model, putting data in and getting a prediction. Now, all of a sudden, the model is the output! Automation development kind of pushes us into a different corner. Instead of doing the tasks, we’re going to be designing the machine that does the tasks. But it still takes a lot of human ingenuity to do these kinds of experiments.
So where do you go with it from here?
The first thing that we did with machine learning at the ALS, and which turned out to quite successful, was heavily based in operations and was rather specific to the ALS — not necessarily something that you could just use anywhere. But we also wanted to follow a more theoretical idea and see if we could use these kinds of tools to assist us in design, so something we’ve been looking at ever since is how we might incorporate machine learning into the design workflow behind a machine like ALS-U.
ALS-U by then had already gone through the design process, so now we’re trying to replicate those results, using a modified tool set that hopefully runs a lot faster and makes it easier for people to find new solutions and makes the whole process more robust. Going forward, another thing I like a lot about this is that the predictive capability allows you to react to something before it actually happens. Hopefully a lot of the things we currently do in accelerators using feedback can be replaced with feedforward. This is very attractive because feedbacks are tricky. They have to be tuned and are geared toward specific modes of operation. Feedforward gets rid of all that and fixes something before it becomes a problem.
You’re a California native. What brought you back to Berkeley?
I was born in Oakland and grew up in Walnut Creek, but then was gone for a long time, so this was like coming back to a lot of childhood memories.
My parents immigrated to the United States from Europe and ended up going back there. I finished high school over there. I missed California a lot (especially the weather!) but the prospect of going to a really great college for what amounts to $1500 a year tuition was kind of a no-brainer. I went to ETH in Zurich and graduated just as the Swiss Light Source was coming up, so that got me hooked on accelerator science at a really exciting time.
After I did a postdoc in the free-electron laser world, I went to work at MAX-Lab in Sweden, where they were building a brand-new fourth-generation storage ring. Once again, that was a really great time — it was the first machine being designed using this new multibend achromat paradigm (which is also a key enabling technology for ALS-U). Being around for that design and commissioning, it was a really great time, and I consider myself tremendously lucky to have hit that window in time.
Several aspects of the Swiss Light Source drew upon experience gained with the ALS, and benefitted from contacts with several of its key people, so quite often, when you wanted to understand how or why something at SLS had been done in a certain way, you’d end up learning about the ALS.
I also met both Dave Robin and John Byrd that way. Dave, of course, is still here at Berkeley Lab as Project Director of ALS-U, and John moved to Argonne to be Accelerator Division director at the Advanced Photon Source, which like the ALS is planning a major upgrade. Anyway, after chatting with John for a while about harmonic cavities, he asked me where I was and what I was doing. After I answered, I remember how he then just blurted out, “Why don’t you come work for us?”. Still makes me laugh to this day. It took me a couple of years to take him up on that.
There aren’t that many large accelerators, so people in the field usually end up traveling around quite a bit. This turns out to be a really great thing. It’s always very enriching to have someone come into the group from outside and make you realize, “There’s a very different way of looking at this problem.”
How are you doing all these team things under pandemic restrictions?
I have to say, it’s really difficult. I haven’t been to the Lab myself since March 16. A year ago I would have said, “We can’t work like that,” but I’m just constantly in contact with people, having all these discussions.
I love sitting in the control room or down at the equipment racks with actual people and looking at things and having face-to-face discussions. That said, it’s surprising how well remote operations have worked, and part of it is because people are trying really hard. I think a lot of us realize that it’s a difficult time, take that into account, and make a real effort to act accordingly — give everybody the benefit of the doubt and realize that they’re all trying to do their best.
It’s mind-boggling how mundane these tools have become to us. It’s become completely normal, but had we faced this pandemic just 20 years ago, it would have been a different story. That’s something to be grateful for.