Cosyne 2024 Roundup: “Compositionality Was The Name of the Day”
The buzz at this year’s Cosyne conference? Compositionality. CDS Assistant Professor of Neural Science and Data Science Cristina Savin, who has been attending the annual Computational and Systems Neuroscience meeting for over a decade, noted this theme across several of the invited talks and workshops.
“In particular, there were interesting new approaches to building compositionality into RL agents,” Savin said, referencing a talk by Alex Pouget from the Geneva University Neurocenter. She also highlighted the work by Laura Driscoll discussed in David Susillo’s invited talk on building compositionality into models of neural circuits.
Another big talk of the meeting was the opening talk in the main meeting, which showcased the intelligence of bees, grounding the conference in the real-world challenges of understanding biological systems. Savin herself presented a poster from her lab on “Learning robust neural representations by straightening natural videos.” The work demonstrated that a biologically-inspired unsupervised learning approach (i.e., straightening) could lead to semantically interpretable representations that are more robust to noise and adversarial attacks than traditional machine learning solutions.
Other highlights? A lively panel on the interplay between neuroscience and AI, which Savin summarized as “the history of the interaction between machine learning and neuroscience and its potential to spur us forward in understanding the brain or building successful AI.” There was also a strong showing from other CDS’ members, with several posters from Professor of Neural Science, Mathematics, Data Science, and Psychology Eero Simoncelli’s group and a contributed talk and posters from Assistant Professor of Neural Science SueYeon Chung’s lab.
Savin, who co-organized a neuroscience workshop in the “I can’t believe it’s not better” tradition established in the machine learning community, has a long history with Cosyne, serving on the program committee for multiple years prior to 2024. The annual gathering brings together experimental and theoretical neuroscientists to exchange ideas and approaches for studying neural systems, a key part of Savin’s own research program that works at the intersection of brains and machines to study learning and memory in neural circuits.
By Stephen Thomas