“We're building an in-silico replica of the brain”
Martin Schrimpf. Credit: Titouan Veuillet (EPFL)
Martin Schrimpf is a new Tenure-Track Assistant Professor at EPFL at the Neuro-X institute. Merging neuroscience with computational modeling, he's on a quest to understand the human brain in computational terms.
What led you into science?
I kind of stumbled into science; it was never the plan. At every academic stage, I thought, “This is it. After this, I'm going into industry.” People suggested I should try science because I was doing well in school. So, I went to a local university in Munich, planning to learn and then work in the industry. During my bachelor's, I founded a startup and worked for another, engaging in cutting-edge work. However, I realized others could do what I was doing, and I wasn't having as much of an impact as I wanted. That led me to dive deeper into machine learning.
So, no training in life science?
No, that came later. My background started with information systems, a blend of computer science and business, which was helpful for startups. I then focused more on software engineering, studying between Munich and Augsburg. Eventually, I wanted more than writing programs; I wanted to have machines learn on their own. That's when I got involved with machine learning. I reached out to the Center for Brains, Minds, and Machines, which led to working on a project with Professor Gabriel Kreiman from Harvard. This experience introduced me to brain science and neuroscience, sparking my interest in the field.
How did you transition from information science to the life sciences?
My first connection to life sciences was through this project. I realized that while many were delving into machine learning, not many were trying to understand the brain using large-scale models. I felt I could make a significant impact in neuroscience, so I decided to focus on understanding how the brain works and did my PhD at MIT’s Brain and Cognitive Science department with Professor Jim DiCarlo.
Can you tell us about your current research?
We aim to understand the brain in computational terms. We're building an in-silico replica of the brain to study it in detail, predict brain diseases, and understand certain mechanisms. We're not just building Artificial Intelligence (AI); we want to create intelligence that functions and behaves like humans. Our focus is on vision and language, studying how neural populations give rise to intelligence in these areas.
Why vision and language?
Vision has been a significant focus in neuroscience, providing ample data for modeling. We also started exploring language about two years ago, finding surprising parallels between vision and language in the human brain. The language system in the brain seems to function similarly to the visual system, providing generically useful features for thought processes.
Can you tell us about Brain-Score?
Brain-Score is a platform we developed to integrate various datasets and benchmarks. Initially, it was a tool I created for my research, but its potential grew as others found it useful. It now hosts over fifty datasets and benchmarks, with hundreds of models submitted by the public. It allows us to find connections between different models and datasets, revealing patterns and insights that might not be evident when studying them in isolation.
What is your philosophy when it comes to conducting scientific research?
I’d describe it as “hardcore empiricism.” While understanding human cognition in intuitive terms is appealing, my primary focus is on building models that work, even if they appear complex. I believe that a functional model of the brain will be very beneficial in the long run, especially when applied to real-world challenges.
Why did you choose EPFL?
EPFL offers a unique blend of strong faculties in machine learning, neuroscience, and neurotechnology. It particularly stands out with its emphasis on entrepreneurship. The university genuinely drives research from the lab into the real world, making it an ideal place for someone like me who's interested in both research and its practical applications.
What are your teaching plans?
I'm still conceptualizing, but the aim is to bridge computer science and life science. We're planning courses that combine these two fields, training the next generation of researchers to think critically and collaboratively. We hope to foster an environment where students from different backgrounds can come together, share their expertise, and work on joint projects.
What do you hope to achieve while at EPFL?
Optimistically, I hope to build an accurate in-silico model of human intelligence and the neural mechanisms that underlie it, particularly in sensory processing and language. Beyond a computational understanding, I believe these models can have practical applications. For instance, if we can stimulate high-level visual cortex using our models, we might be able to restore some level of vision for individuals with visual impairments.