2024 GAC 2
Attending to errors in predictive coding: a collaborative community experiment through the OpenScope program
Jerome Lecoq, Allen Institute
Michael Berry, Princeton University
Colleen Gillon, Imperial College London
Konrad Kording, University of Pennsylvania
Short description
Predictive coding stands as a leading theory of cortical function, holding significant implications for our understanding of perception, cognition, learning, and overall brain function. Multiple hypotheses have vied to explain how the cortex performs predictive coding. Central to all these proposed explanations is the following question: What mechanism(s) underlie the subtraction of predictions from input sensory data? While many answers to this question have been proposed, two competing hypotheses have come to the forefront: a cellular hypothesis and a dendritic hypothesis. Our workshop unites experimentalists and theorists with complementary and conflicting results and hypotheses to collaboratively design a neuroscience experiment addressing this question. Our proposal is strengthened by collaboration with the Allen Institute’s OpenScope Program, an NIH-funded brain observatory ready to perform the proposed experiment on the Allen Institute’s pipelines—without the need to wait for a grant application cycle (http://openscope.ai). Datasets stemming from this endeavor will be made openly accessible to the entire scientific community for analysis.
Schedule
5.15 pm - 7.00 pm, Sala de Puerto Rico
5.15 pm - 5.45 pm
Session 1. Presentation of Theories and Experimental Predictions (30 min)
Speakers will provide broader context about predictive coding and lay out a specific prediction about how subtractions are implemented in the brain. Each talk will be 8 min with 2 min for directed questions.
Michael Berry, Princeton University. What is predictive coding for?
Claudia Clopath, Imperial College London. Prediction-error neurons in circuits.
Fabian Mikulasch, Max Planck Institute, Göttingen. Dendritic predictive coding theory of hierarchical inference in cortex.
5.45 pm - 6.15 pm
Session 2. Comments from Experimentalists (30 min)
Speakers will present key experimental data taken in their lab and will comment on how their data impacts the larger issues raised in session 1. Each talk will be 8 min with 2 min for directed questions.
Colleen Gillon, Imperial College London. Distinct signals in apical dendrites and cell body populations.
Jeff Gavornik, Boston University. When and where of expectation violations: complexity and limitations of visual predictive coding.
Jordan Hamm, Georgia Institute of Technology. Feedback modulation of V1 supports predictive processing through a neuron-type specific circuit.
6.15 pm - 7.00 pm
Session 3. Discussion (45 min)
In order to seed discussion with some concrete proposals, three speakers will each present an experiment to address the predictive mechanisms on one slide only. Each presentation should last no more than 3 min, followed by 2 min for specific questions. After all specific proposals, we will open the floor for general discussion for the remaining 30 min. All workshop participants will be polled after the workshop to make further comments and to vote on their preferred experimental proposal.
Jerome Lecoq, The Allen Institute. Are all errors the same?
Colleen Gillon, Imperial College London. What signals do we find in layer 1?
Michael Berry, Princeton University. Is E-to-E synaptic plasticity intrinsic to predictive coding?
Speakers
Michael Berry
Claudia Clopath
Colleen Gillon
Jeff Gavornik
Jordan Hamm
Jerome Lecoq
Konrad Kording
Comment on this GAC proposal!
Instructions for CCN community members:
Download the proposal PDF using the button above.
Use the Disqus box below to leave comments and reviews, and discuss the submission. You'll need to create a Disqus account or log into Disqus using a google account or similar in order to leave a comment/review.