when Eva DyerAssociate Professor of Bioengineering and Computer and Information Science at Rachleff, talks about the brain, she doesn't just speak like a neuroscientist. She speaks to the rhythm of someone who listens deeply. Once a jazz singer and multi-instrumentalist, Dyer coordinates another kind of harmony. With the help of artificial intelligence, it is to find hidden signals in the brain.
“As a child, I didn't really touch on what scientists were or what they looked like,” says Dyer. “I was the first person in my family to receive a higher education. But I didn't become a scientist. I followed a question that interested me.”
One of those early questions was how to interpret music.
In high school, Dyer was immersed in music. She sang jazz, played piano and drums, and experimented with many other instruments. At the same time in her physics class she was learning about sound and vibration.
“To be able to explain what physics and mathematics love most has blown my mind,” she says. “It was magical to realize that sound, emotions, science wasn't separate, but was deeply connected.”
That connection became her compass. Dyer majored in Audio Engineering as an undergraduate at the University of Miami. The school is an interdisciplinary major across schools with dual degrees from both engineering and music schools. To that extent, she thought she might become an acoustic consultant who helps to design speakers and architecture to enhance sound and music. But a single concept – the perception of hearing, or the way our brains can hear in the world around us – has changed everything.
“I realized that the brain is an important part of how you experience sounds,” she says. “It led me to neuroscience.”
From there, Dyer followed her instincts and data. She leaned on machine learning and signal processing to develop a computational tool to analyze how the brain responds to complex environments. This was before people began to join together in the fields of AI and neuroscience, but they were able to see where interdisciplinary possibilities were heading.
“When I first started graduate school, I wanted to use machine learning to understand the brain, but at the time there was no gathering of data science and neuroscience at the time,” she recalls. “Building the foundations of applied mathematics and signal processing was an important stepping stone for me. By the time I completed my PhD in 2014, new high-resolution methods were generating large datasets. It was an exciting moment to bring deep learning to neuroscience. It's amazing to see how far things have come.”
Today, Penn's Dier Lab is at the intersection of neuroscience and AI. She works closely across pen medicine and disciplines to use powerful machine learning techniques to decode brain signals, including everything from intentions and movements to mental health symptoms.
“With a brain computer interface, you are listening to the brain and trying to translate your thoughts into actions. This could help someone living with paralysis regain control over their movements and voices,” explains Dyer. “But we know that we need large data to do that well. With the tools used in large language models, we can pre-train AI systems on large amounts of neuroscience data, and we can tweak them to extract information from a small subset of data and tap on information that we don't even know is there.”
In other words, By scaling up and training with more diverse data sets Beyond a variety of animals, experimental setups, and various mental states, Dyers Lab has discovered that it can derive new training data that can collect information from new brains more quickly and effectively.
One of her latest projects, supported by Hypothesis FundCatalytic Seed Grants, which supports bold, early-stage scientific research aimed at addressing systemic risks to human and planetary health, are particularly ambitious. Map brain diversity at the level of individual cell types. The idea here is to use AI to generate a “description” of the algorithm for each cell. This is the type of code that captures what each of them does in the system, and doesn't have complicated words yet.
“We can't really explain neuronal function in natural language alone, especially using all the complex functions in our brain,” says Dyer. “But we can probably explain it in code. Our hope is that a large-scale language model trained for programming will help us write algorithms that explain the behavior of different neurons in our brain. With the help of this seed fund, we are building a new language that speaks about the brain.”
The language hopes that researchers will help them understand which cells are affected in diseases like Alzheimer's, or help them design more accurate interventions for disorders like OCD.
But for Dyer, none of this is about reducing the brain or AI to a simple formula.
“AI and the brain are both black boxes in many ways,” she says. “It's fascinating to think we open them and see exactly how they work, but the truth is even more complicated. The exciting thing is that when we start to expand, we find signals that we didn't know existed in our data. That means we need new ways to represent our brain datasets and new ways to ask questions from these data.”
And the questions are getting more complicated.
“In science, it can be attractive to control everything in an experiment,” Dyer points out. “But it often eliminates the richness of the real system, especially in the brain. I accept this complexitydon't avoid it. ”
That spirit fits well with the school's deep commitment to interdisciplinary research, its proximity to pen medicine, and the pen engineering engineered to the vibrant, walkable community she currently calls home.
“Pen is a place that encourages interdisciplinaryity, both through culture and the physical location of the schools I work with. We are all within a few blocks of each other,” says Dyer. “There is real momentum around AI with new majors, new faculty and centers, and Penn is investing in infrastructure, talent and space to do this right.
That includes cultivating the next generation of thinkers and builders. This November, Dyer organized a Buildson aimed at researchers who are passionate about using AI to explore the brain. This event brings together AI and Neuroscience developers to explore and contribute to Dyer and her team's new open source software packages Torch Brainfor designing and scaling AI models. Their goal is to build standards and benchmarks that drive the field forward, and create new tools such as brain machine interfaces, neural data analysis, and more. Please see details. Visit the tutorial here.
Visiting her will help you learn more about her work, just as you would like to work with Dyer for potential collaborators, graduate students and mentees. Research Website.
