Understanding neural computation in natural environments: A conversation with Siwei Wang

The NSF-Simons National Institute for Theory and Mathematics in Biology is composed of investigators at the forefront of innovative research at the interface of mathematics and biology. NSF-Simons NITMB Affiliate Members each bring unique perspectives vital for developing new mathematics and inspiring biological discovery. One such NITMB Affiliate Member expanding our understanding of collective cell behavior is Siwei Wang. 

Siwei Wang, Assistant Professor, Department of Neurobiology and Behavior, Stony Brook University 

Siwei Wang is a tenure-track Assistant Professor in the Department of Neurobiology and Behavior at Stony Brook University. Professor Wang uses information theory, Bayesian inference, and representation learning methods to investigate how neural representation evolves along the sensory-to-behavior arc to enable effective future action planning based on past experiences. 

We spoke with Siwei Wang to explore her work with neural computation and explore potential applications of her research in medical fields and in artificial intelligence. 

What is a big question you’ve been asking throughout your research? 

“I started my journey in theoretical neuroscience to really think about why neural computation works in its natural environment, and the flip side of that, why evolution sculpts neural systems and neural computation the way it is. The central question is if there is a canonical operation that’s shared across species, if what we learn from one animal is transferable to another, and later can help us understand the complicated human brain.” 

What disciplines does your research integrate? 

“It’s still evolving because I’m only eighteen months into this professor job. Before, I was a postdoc at the University of Chicago. At the time, I mainly worked with information theory and some of the interaction with statistical mechanics and the physics angle. In my new research, I’m bringing in a bit more machine learning. If we obtain insights from canonical computation shared across species, we also take those as inspiration to build AI systems that are more efficient. Energy efficiency is still the biggest problem in AI. Our brain operates at 19 watts, the same as a lightbulb, while most large language models require thousands more.” 

Where do you find inspiration? 

“If I look through how I choose my research topics, the inspiration is human behavior. My PhD was actually in computer science for speech and language processing. The reason I chose that was that my grandfather lost his hearing because of a neural disease before I joined the PhD program. At the time, I thought building software would help patients like that. Later, I went into computational neuroscience and started to look at how neural circuits compute for operation and survival-related tasks like chasing a mate, escaping from predators, or catching food. For studying maneuvers like walking and flying, I took inspiration from myself doing sports. There was a crazy phase of my life where I was a skydiver, so I had this experience where I knew moving around different mediums has different levels of challenges or computation you’re facing. Now I’m also working on the brain-computer interface and cross-species comparison of dexterity. That comes back to how our brain connects to our body. We want to help human patients while taking inspiration from animal recordings.” 

What aspects of your work could be interesting to mathematicians or applied to biology? 

“Mathematically, we are essentially going into this big umbrella of questions of dimensionality reduction. One example is when we are doing human-computer interaction. When trying to help a pathologically paralyzed patient, we are looking at their imagination and at their motor cortex to see how they believe they are grabbing or doing a movement, while the projection pathway from their brain to their muscle is broken. Even if you cannot seek feedback from the external world, once this behavior is ingrained in your brain, you still have a representation there that you can retrieve. What is the way this representation becomes so persistent in your brain, given you can still do that after years of paralysis? For example, my grandfather, even after tens of years unable to hear anything, could still talk normally. That means we want to represent the whole behavior of talking or movement in some energy-efficient or robust way where your brain can still retrieve it without checking the external world. Operation-wise it’s dimensionality reduction. We want to find the most compressed form that is easiest for your brain to store. And mathematically, another thing we are looking at is information processing in a highly complex system. We want to find the common latent state shared between mediums. In the most ideal scenario, biologically, this would inspire new experiments or clinical trials.” 

What about the NITMB do you find exciting? 

“NITMB helps me to explore the applications of mathematics to biological systems in my work, and how mathematics is engaging different but relevant biological systems, providing inspiration. If you share the same statistical or mathematical structure, if we find problems that work in different ways, maybe we can deliver novel solutions in those as well. NITMB encourages us to find common elements of mathematical solutions that help us know which biological principles exist behind them.” 

What career achievement are you most proud of? 

“It’s cultivating a mindset. When I was a postdoc, I was lucky to identify principles in neural processing based on small animal models which were later demonstrated working in humans. However, the process involves a lot of doubt and questioning, both from myself and my peer reviewers. While I understand thinking out of the box necessarily invites disbelief in its beginning, my skin was not thick enough in some scenarios. Scientific discovery is a two way street. all publications are meant to invite folks to check if my conclusion makes sense given the evidence I presented. But I recently read an interview with prominent economist Daniel Kahneman, and one thing people say about him is that his face lights up when people say, ‘your work is wrong.’ So, I hope a couple of years down the road I have that attitude that if somebody’s feedback thinks outside of the box I constructed, that’s a great thing.” 

Outside of your research, what other interests do you have? 

“I play a lot of sports. I’m mostly interested in outdoor activities like kayaking and rock climbing. I would say rock climbing is what I like the most. I also enjoy biking and solving puzzles.” 

What are you hoping to work on in the future? 

“I am proud to work with The Maclean lab and the Bensmaia lab of University of Chicago on natural behaviors, one in mice and one in monkeys and humans. I am connecting the dots to see how we compare the representation of natural behavior across species. I am also working with the Gregory Schwartz group at Northwestern, looking at cells from the retina to the primary visual cortex. Even in the current stage, we know a lot more about the brain, but we still do not have the ground truth of what these cells look like. We are hoping these collaborations will lead to applications in AI systems. We want to deliver insights to explain why behaviors we find in mice are similar to those in monkeys. Explainable AI is one of the inspirations. I will also be looking forward to working with an NITMB Fellow that’s interested in the cusp of information theory, representation in machine learning, and perturbation theory from statistical mechanics for any of the mathematical foundations of the neuro-AI work I am developing. And as a newly affiliated NITMB member, I look forward to establishing collaborations within NITMB through visits for research and later organizing conferences and workshops with my collaborators” 

More information on Siwei Wang’s work is available on Google Scholar.