Networking for the next generation of neuroscientists
When the pandemic shut down normal day to day operations in 2020, UCI cognitive scientist Megan Peters pivoted her planned summer school activities with colleagues into a multinational, multifaceted neuroscience networking, mentoring and education program. Neuromatch Academy - which now boasts more than 8,500 alumni - is an online computational neuroscience and deep learning summer school that offers an intensive and creative three-week, six-hours-a-day learning experience intended to presage and augment activities and concepts students and professionals will encounter in their studies or jobs.
Launched with far-flung partners Konrad Kording at the University of Pennsylvania, Gunnar Blohm at Queen's University in Ontario, Canada, Paul Schrater at the University of Minnesota, Sean Escola at Columbia University and Brad Wyble at Penn State, the academy uses algorithmic matching—much like a dating app profile would—to match participants including undergraduates, graduate students, postdocs and industry professionals from around the world into online pods or groups.
Over the past three years, Peters has helped set up the academy’s corporate and financial infrastructure, established its non-profit status, helped form and guide an operating/executive committee and sub-committees responsible for curriculum design, and had a hand in a host of other duties that doubtless make her glad that these days she’s able to focus more on long term vision and mission rather than the day-to-day operations.
Like her co-founders, the academy is Peters’ side gig. Her chief vocation is teaching and conducting research as an assistant professor of cognitive sciences in UCI’s School Social Sciences. She recently found the time to speak with us about the academy and her research at UCI.
Would you tell us what’s behind the Neuromatch name?
The idea is to let people know this isn't just any online program. We’re drawing upon the best tools we have available to bring people together and create a sense of community. We take the algorithmic matching to extremes. Because the academy has participants in 120 countries, we match people by time zone so they can all be awake at the same time. We match by the general educational level they are at, so we’re not necessarily mixing undergraduates with postdocs, because there are probably differences in background and different expectations about what to get out of the school.
We also ask, “What language do you speak?” The course material is in English, but many participants would prefer to speak with their podmates and TA in Spanish, Mandarin or a host of other languages. We ask, “What are your interests within neuroscience? Are you interested in electrophysiology and fMRI (functional magnetic resonance imaging)? Are you interested in modeling? Are you interested in cellular neuroscience? Do you want to do a group project? Which data set are you most interested in working with?”
What made you all decide to launch the Academy?
The need for it was catalyzed by the onset of the pandemic. I had been planning with Konrad, Gunner and Paul to continue the in-person summer school they had run for 10 years. When the pandemic hit, we clearly couldn't do that. So what else could we do?
Once we decided to run it online, we quickly realized that, given the amount of work needed to put it online, we would be remiss if we ran it for only 50 people, and that we had an opportunity to create something that would reach a much larger and more worldwide audience, in a way that would democratize access to this type of educational opportunity, regardless of your location or financial means.
There was a need for that because these summer schools have such a profound impact on young scientists’ lives, in terms of the education and networking they get, and for the resources they come away with. In 2013 when I was a Ph.D. student, I attended the Computational Sensory Motor (CoSMo) summer school Konrad, Gunner and Paul ran, and it changed my life, not just in terms of the education I got, but also the science friends I came away with.
Can you give some idea of the scope of this enterprise?
From our three years of summer school, I think we have about 8,500 alumni now. To run the back end of things, we have 250 to 300 volunteers. We have about 500 teaching assistants, who hail from maybe 100 countries. For the group projects, students have hundreds of mentors we’ve matched with them. The mentors lend their expertise, for example, to help groups to develop their project or even turn it into something that could be presented at a conference or turned into a paper.
What are some of the things the Neuromatch sessions cover?
There are two courses in the academy. One is Computational Neuroscience, which covers everything from “What is a model?” to “How do I build a model of the nervous system?” to “How do I build a model of the mind?” all the way through some of the most exciting techniques out there right now, like Bayesian approaches to studying evidence-based decision making, accumulation reinforcement learning--that kind of stuff plus some theoretical computational neuroscience as well.
Then there is the Deep Learning course, which covers what we think is the state of the art in deep learning right now. That's covering the basics like autoencoders all the way through transformers. I don't think we have stable diffusion models in there yet, because they just came out.
For a non-scientist, probably the easiest way to look at deep learning would be to take the idea of a deep neural network, or a deep convolutional neural network, where you have an input—let’s say an image--and the goal of the network is to determine, “Is this a cat or a dog or a toaster?” based on connections between the input and the pixel space. It takes a little square of pixels, then another, then another, then takes an average of those into the next layer up, which has fewer nodes in the network. It continuously takes tiny spatial averages across the entire scene. And at the top level you have a very simple network: cat, dog, toaster. Whichever one of those nodes has the highest activity level, that's the choice the network's going to make. And so then you say here's an image of a cat, and you train the weights of those connections as averages across all the layers and you say, “Did you pick cat, dog, or toaster?” If it gets it right, you say, “Great, good job; keep those weights.” If it gets it wrong, you back-propagate the error and maybe update the weights differently. If you do that maybe 100,000 times over, you get a system that can see a new image and determine that it is definitely a dog and not a toaster.
And there are tons of applications for deep learning. You want your Tesla to be able to know whether the thing speeding towards it is a child or a car? It's going to use some variant of that learning process, as is having a machine trained to look at different x-ray images and decide whether a fuzzy looking blob is lung cancer or not. Machine learning is also driving Netflix and YouTube recommendation algorithms, assuming “If you like this, then you’ll like these things.”
What initially sparked your interest in neuroscience, and how did that lead to where you are now?
I guess it started back in high school when I became fascinated with the idea that our brains are wholly responsible for the construction of our experiences about the universe, and about our interactions with other people, like the explanation for why it feels a certain way to interact with someone; the explanation for why red looks red to us. The explanation for all that stuff lies in our heads, in the hardware and the software of the mind.
I started looking at these things from the philosophical side, but then I also became fascinated with physics and the idea that we could explain the physical universe with math. I also wanted to be able to explain the differences in experiences between people, the idea of subjective experience of reality.
And then I got to college and found neuroscience, psychology and cognitive science, and that scratched both itches, where I could do the philosophy and the understanding of the mind from the software side, and also have the quantitative rigor to find explanations for what was going on. So it was always that drive to understand how the brain constructs our subjective experiences of the world. That kind of carried me through.
How did pursuing that lead you to UCI?
When I majored in cognitive science at Brown University, I really loved being able to ask complicated questions about the brain and mind in a quantitative way, while also having strong influences from philosophy. After doing my Ph.D. at UCLA in psychology with a focus on cognitive and computational neuroscience, I was a postdoc there with Hakwan Lau for a few years, and then took a position as assistant professor in the bioengineering department at UC Riverside as part of a program aiming to integrate bioengineering, neuroscience and psychology.
I was happy at UCR and had fantastic colleagues, but when I was invited to apply at UC Irvine in cognitive sciences, I couldn’t pass up the opportunity to come home to cog sci and be in a department where my students would have access to targeted training and to other faculty who “spoke my language” so to speak.
What I like about UCI is exactly what I thought I would: the department’s educational goals and offerings are a really fantastic fit for my own research program and for training my students. I always think we could develop more courses at the graduate level, I suppose, but that takes time and resources and already we have a lot of targeted classes that my students really can take advantage of. I have already developed strong collaborations with several faculty in my department, and really enjoy learning about their students’ research as I sit on their committees as well. I also like the large and healthy neuroscience-related community on campus, in part facilitated by the Center for the Neurobiology of Learning and Memory; through events at the CNLM and elsewhere, I’ve made connections with faculty in neurobiology and anatomy/physiology, among others, and it’s very convenient that many of us live in University Hills so it’s easy to get together — even during COVID.
What manner of things does your research at UCI explore?
Using computational cognitive neuroscience and computational modeling, we're looking at things like uncertainty, metacognition and introspection. For example, we're looking at what the brain is doing on a daily, hourly, minute-wise basis: It's taking in information from the world and extracting statistical regularities in that information so that you can understand what you're looking at or what you're hearing. And so, the fact that you can understand language has to do with the fact that you have paired the acoustic properties of the sound waves that I am producing with meaning, through decades of experience with statistical patterns and the fact that you are deciding that the sound comes from my mouth and not over there. It has to do with the congruence between your visual signal saying ‘There's a thing happening here’ and your auditory system agreeing ‘There's a thing happening here.’ And if I was a ventriloquist, I could potentially throw my voice and I could hijack that simultaneity of visual and auditory signal to make you think the voice was coming from over there. And the reason I can do that is because your brain is coming up with the most likely explanation for the patterns it's receiving, so that it can use those to better predict what's going to happen next, so that, from an evolutionary perspective, you can learn to navigate your world, learn to not get eaten by other creatures and so on.
What are some of the factors that distinguish the human mind from other creatures’ brains? What makes us human?
Scientists will tell you it’s the prefrontal cortex, which is over-developed in humans relative to every other species. We think of it as the seat of reason, of planning, of executive control, working memory, problem-solving and metacognition, due to introspection.
For example, what we were just talking about is the idea that you can reason your way through to what might happen next. So if you did something, you can run an internal stimulation of the world going forward from that, and then pick the best course of action, and then reflect on your capacity to do that, or reflect on your capacity to act in the world in a rational and goal-directed manner. Those abilities are, we think, more developed in humans relative to other species. We also are the only species that has language in a true syntactical format. We can teach chimps and gorillas to use symbolic reasoning, but they don't ever quite learn syntax, or at least not like we can use it. I don't know whether languages are a factor in making us human, but it certainly is something that makes us stand apart from other species.
It’s a busy enough life being a professor. Why did you determine to take on doing the work involved with the Neuromatch Academy?
Because I think it is important. I get a little time every year that I can spend on volunteering. As a member of the UC system, you have to fill out a conflict-of-commitment form every year about how you spend your off time, basically in activities that are related to your job. You have a little allotment of time that you can fill every year, and this feels like the most important way I can spend those hours.
Applications for summer 2023 open March 31. Learn more about becoming part of the next cohort.
-Jim Washburn for UCI Social Sciences