On November 12 and 13, 300 experts in the biology and computation of the brain gathered at the National Institutes of Health campus in Bethesda, MD. Online, there were as many as 1,200 listening in. This was the 2024 BRAIN NeuroAI Workshop, organized by the BRAIN Initiative, a public-private partnership accelerating neurotechnology advances with significant funding - $402 million from Congress in the current fiscal year.

Carney Institute affiliate Carina Curto, a professor of applied mathematics and brain science, was in the audience as a presenter and panelist. She shared the stage with luminaries that included computational neurobiologist Terry Sejnowski, computational neuroscientists Frances Chance and James Bradley Aimone, engineers Kwabena Boahen and Mitra Hartmann, and neuroscientists Tony Zador, Andreas Tolias and Doris Tsao.

The meeting was designed to highlight this talent through rapid-fire talks and moderated discussions. Speakers were organized into four sessions, and each had just eight minutes to present their ideas. 

Curto’s fast pitch: With inspiration from biology, and understanding from mathematics, researchers in the fast-growing field of NeuroAI can build entirely new artificial neural networks - networks that work with less data and run with less electricity.

Curto said that neuroscience concepts like neuromodulation, rhythms and oscillations and dendritic computation can now be studied in fine detail in the brains of small creatures like fruit flies. In a landmark October 2, 2024 issue of the journal Nature, scientists published a neuron-by-neuron map of the entire brain of an adult fruit fly. Insights from this science, Curto said, can be used to design entirely new AI systems. 

Right now, AI systems are built on very large deep neural networks trained on very large data sets that require very large amounts of electricity. A 2024 report by Goldman Sachs Research shows that AI growth will cause worldwide data center power demand to grow 160% by 2030 – a once-in-a-generation increase. Curto said many workshop participants were motivated to devise more energy-efficient, climate-smart AI systems.

While neuroscience can provide inspiration for innovation, Curto said, progress will also be powered by mathematics. 

“Math tells us how to train these networks,” she said. “A deep neural network is basically a family of functions with billions or trillions of parameters. You want a network to do something, but you first need to find the right function. That’s the training process. And we don’t know the best way to do the training. We’re not doing it efficiently - but math can help with that.”

In her workshop presentation, Curto talked about threshold-linear recurrent networks, a current focus of her work that’s backed by BRAIN Initiative funding. These network models are easier to control mathematically, and can result in more elegant, efficient AI systems. 

As an example, Curto shared new work from Juliana Londono Alvarez, a Brown postdoctoral research associate in applied math, who used a recurrent model to encode multiple gaits from bounding to trotting, in multiple animals, from horses to squirrels. Londono Alvarez presented a poster of her findings at the workshop.

Curto’s big takeaway from the meeting: there’s a big wave of enthusiasm building for the budding field of NeuroAI.

“There were a lot of different scientists, from different fields and institutions, at the meeting and a significant number of funding agencies were represented, from NIH to NASA to DARPA,” she said. “A lot of people want better AI systems, whether it’s for robotics or medicine or space exploration. I’ve believed for a long time that we can develop better AI with new insights from neuroscience and mathematics. It was eye-opening to see just how many other people believe it. It made me optimistic about the future of this field.”