NeuroAI Takes the Stage at Kempner Institute Symposium
At the Frontiers in NeuroAI symposium, the Kempner Institute brought together researchers who are defining a new scientific field
A two-day symposium, held both in-person at Harvard, and virtually through a live stream, brought together more than 1,000 researchers and trainees from 37 countries around the world.
Photos: Anna Olivella
The early years after the emergence of a new field are a dynamic and transformative time, when a new community coalesces around a set of scientific questions and a common vision for how to answer them. The field of NeuroAI, which brings together neuroscience and artificial intelligence, is currently in this period of expansion and bloom.
Last week, the Kempner Institute organized a showcase for this growing field: the Frontiers in NeuroAI symposium. This two-day gathering brought together more than 1,000 researchers and trainees from 37 countries around the world. Participants from a variety of disciplinary backgrounds— including neuroscience, psychology, machine learning, computer science, physics, robotics, and psychology — convened to share the newest data and analytical techniques related to the study of brains and their artificial counterparts.
The symposium, which took place at Harvard’s Science and Engineering Complex in Allston, MA, as well as virtually via a livestream feed, occasioned a kind of conversation between forms of intelligence: on the one side the brain, which enables biological intelligence, and on the other, a family of modern AI methods including artificial neural networks (ANNs).
The symposium highlighted several threads of inquiry within NeuroAI including: new frameworks for studying brains and AI models; AI-powered insights into the brain; next-generation AI systems; and a deeper understanding of current AI models.
Over the course of two intensive days of talks, poster presentations, and impromptu brain-storming sessions, students, postdocs, and senior researchers took stock of how far the fledgling discipline of NeuroAI has come in just a few years. Conversations built on each other, sowing the seeds for new ideas and collaborations.
For anyone looking for an example of NeuroAI in action, the Frontiers in NeuroAI symposium offered a compelling snapshot of the field. With its burgeoning roster of linked research programs, the NeuroAI community is building a positive feedback loop that benefits both technological development and neuroscientific research. The Frontiers in NeuroAI symposium was proof positive that this feedback loop is already stimulating innovation and scientific discovery.
Watch the full recordings
The full library of recordings of expert talks from Frontiers in NeuroAI are available on the Kempner Institute website. Learn more about subjects of individual talks and access talk links below.
New frameworks for studying brains and AI models
- SueYeon Chung explained how physics & geometry can illuminate how the brain computes with neural manifolds. Watch: Computing with Neural Manifolds
- Cengiz Pehlevan showed how summary statistics offer a window into neural network representations, and how they change during learning. Watch: Summary Statistics of Learning Link Changing Neural Representations to Behavior
- Mozes Jacobs showed that traveling waves can enable neural networks to ‘hear’ the shapes of objects. Watch: Can Your Neurons Hear the Shape of an Object?
- Veronica Chelu showed how excitatory-inhibitory dynamics in recurrent neural networks can affect decision making and reinforcement learning. Watch: Excitatory-Inhibitory Dynamics in Adaptive Decision-Making
AI-Powered insights into the brain
- Surya Ganguli showed how recurrent neural networks can teach us about dynamics of neurons in the brain’s entorhinal cortex, which contributes to reasoning and imagination. Watch: Theories of Learning, Imagination and Reasoning: of Mice and Machines
- Karel Svoboda demonstrated how brain-computer interfaces can shed light on synaptic learning rules. Watch: Illuminating Synaptic Learning
- Mark Andermann brought experimental data and computational modeling together to understand cortical reactivations. Watch: Offline Cortical Reactivations of Recent Experiences
- George Alvarez showed how long-range feedback projections can enhance the alignment between ML models and human visual processing. Watch: Emergent Cognitive & Neural Alignment Between Biological and Artificial Systems
- Konrad Körding explained why causality is so difficult to uncover in neuroscience, and how it could still be identified in some situations. Watch: Causality: Why Most Claims to Causality are Bogus
- Sara Matias showed evidence that the striatal dopamine circuit can mediate distributional reinforcement learning. Watch: Broadly-Projecting Mesolimbic Dopamine Neurons
- Mackenzie Mathis demonstrated work on developing “foundation models” for behavior, unifying the analysis of brain data and animal movement data. [Not recorded.]
The next generation of AI systems
- João Sacramento discussed how transformers & related architectures enable in-context learning of sequence data. Watch: Sequence Prediction through Local Learning
- Yilun Du explained how optimizing energy functions can help solve challenging navigation & reasoning problems. Watch: Learning Compositional Models of the World
- Luke Zettlemoyer walked through different approaches to building multimodal foundation models. Watch: Mixed-modal Language Modeling
Deeper understanding of AI models
- Asma Ghandeharounian demonstrated techniques to enhance interpretability of large language models, including the Patchscopes framework. Watch: Model Interpretability: from Illusions to Opportunities
- Ellie Pavlick explained how compositionality emerges in large language models. Watch: What Came First, the Sum or the Parts?
- Fernanda Viegas looked under the hood of AI chatbots to uncover implicit inferences about the user. Watch: What Do AI Chatbots Think About Us?
- Mason Kamb presented a predictive theory of combinatorial creativity in diffusion models. Watch: An Analytic Theory of Creativity in Convolutional Diffusion Models
About the Kempner Institute
The Kempner Institute seeks to understand the basis of intelligence in natural and artificial systems by recruiting and training future generations of researchers to study intelligence from biological, cognitive, engineering, and computational perspectives. Its bold premise is that the fields of natural and artificial intelligence are intimately interconnected; the next generation of artificial intelligence (AI) will require the same principles that our brains use for fast, flexible natural reasoning, and understanding how our brains compute and reason can be elucidated by theories developed for AI. Join the Kempner mailing list to learn more, and to receive updates and news.
PRESS CONTACT:
