COSYNE Workshop on “The geometry & dynamics of learning: Bridging analytical and experimental insights into neural representations”
Recent advances in neural recording have revolutionised our ability to record from populations of neurons over time and at higher resolution. These advances stand to challenge existing knowledge about the dynamical nature of learning in neural systems, which, in turn, may reveal how representations underlying complex behaviours develop over time in neural architectures. Simultaneously, progress in artificial neural networks has revealed the geometric properties of emergent neural representations, along with analytic insights into the learning dynamics that shape these representations. Despite such massive strides in both fields, we still lack a common understanding of how distinct learning strategies in the brain can shape the observed neural population structures. The present workshop aims to bridge that gap by fostering discussion between experimentalists working on changes in neural activity in the brain over learning and theorists working on learning in neural network models in order to better understand learning at the level of populations of neurons.
Speakers
Imperial College London
Technion
Harvard University
Princeton University
École Normale Supérieur
University of Edinburgh
Imperial College London
Allen Institute
Stanford University
Flatiron Institute
University College London
ETH Zürich
Champalimaud Research
Queens University
Janelia
Harvard University
UC San Francisco
Stanford & Oxford
Schedule
Start | End | Session | Speaker | Title |
---|---|---|---|---|
Monday, March 4th: Morning 1 | ||||
9:30 AM | 9:35 AM | opening remarks | Erin Grant & Stefano Sarao Mannelli | |
9:35 AM | 10:00 AM | invited talk | Juan Álvaro Gallego | Circuit properties and experience shape neural population structure and influence learning |
10:00 AM | 10:25 AM | invited talk | Valerio Mante | What do neural representations say about computation, and what not? |
10:25 AM | 10:50 AM | invited talk | Tim Buschman | Learning to control cognition |
10:50 AM | 11:10 AM | coffee break | ||
11:10 AM | 11:35 AM | invited talk | Jenelle Feather | Neural population geometry of category learning |
11:35 AM | 12:00 PM | invited talk | James Whittington | Relating frontal and hippocampal algorithms for sequence memory |
Monday, March 4th: Afternoon 1 | ||||
3:30 PM | 3:35 PM | opening remarks | ||
3:35 PM | 4:00 PM | invited talk | Cengiz Pehlevan | Dynamics of representational drift |
4:00 PM | 4:25 PM | invited talk | Omri Barak | How learning shapes representations |
4:25 PM | 4:50 PM | invited talk | Emily Oby | Dynamical constraints on neural population activity |
4:50 PM | 5:10 PM | coffee break | ||
5:10 PM | 5:35 PM | invited talk | Marius Pachitariu | Unsupervised pretraining in biological neural networks |
5:35 PM | 6:30 PM | panel | Omri Barak, Tim Buschman, Jenelle Feather, Valerio Mante, Emily Oby, Marius Pachitariu, Cengiz Pehlevan, James Whittington | #cosyne-geo-dyn on Sli.do |
Tuesday, March 5th: Morning 2 | ||||
9:30 AM | 9:35 AM | opening remarks | ||
9:35 AM | 10:00 AM | organizer talk | Alex Cayco Gajic | Identifying learning-induced changes in neural dynamics with low tensor rank recurrent neural networks |
10:00 AM | 10:25 AM | invited talk | Lea Duncker | Low-dimensional neural population dynamics for robustness and learning |
10:25 AM | 10:50 AM | invited talk | Claudia Clopath | Drift of neural ensembles driven by slow fluctuations of intrinsic excitability |
10:50 AM | 11:10 AM | coffee break | ||
11:10 AM | 11:35 AM | invited talk | Kenneth Harris | Learning orthogonalizes visual cortical population codes |
11:35 AM | 12:00 PM | organizer talk | Angus Chadwick | Learning reconfigures cortical dynamics to enhance representations for behaviourally-relevant stimuli |
12:00 PM | 12:30 PM | panel | Blake Bordelon, Claudia Clopath, Lea Duncker, Kenneth Harris | #cosyne-geo-dyn on Sli.do |
Tuesday, March 5th: Afternoon 2 | ||||
3:30 PM | 3:35 PM | opening remarks | ||
3:35 PM | 4:00 PM | invited talk | Blake Bordelon | Mean field theory of representation learning dynamics |
4:00 PM | 4:25 PM | invited talk | Francesca Mastrogiuseppe | Quality of the internal representation determines learning performance in recurrent neural networks |
4:50 PM | 5:10 PM | invited talk | Kayvon Daie | Local circuit plasticity in motor cortex during learning |
4:50 PM | 5:10 PM | coffee break | ||
5:10 PM | 5:35 PM | invited talk | Xulu Sun | The geometry of cortical neural population dynamics to support motor learning and motor control |
5:35 PM | 6:30 PM | panel | Alex Cayco Gajic, Angus Chadwick, Claudia Clopath, Kayvon Daie, Lea Duncker, Kenneth Harris, Xulu Sun | #cosyne-geo-dyn on Sli.do |
Organizers
École Normale Supérieur
University of Edinburgh
McGill University & Mila
University College London
University of Washington
University of Osnabrück
University of Edinburgh
McGill University & Mila
University College London