|
Together, apart: how invariances and selectivities contribute to predictiveprocessing| title | Together, apart: how invariances and selectivities contribute to predictiveprocessing |
|---|
| start_date | 2022/09/19 |
|---|
| schedule | 12h UTC +1 |
|---|
| online | yes |
|---|
| visio | https://us02web.zoom.us/j/83888993899?pwd=ZFZ0STdQVmhESkI2bnVabWRWV0RpQT09 |
|---|
| location_info | NeuroSpin's amphitheater |
|---|
| details | En présentiel et zoom |
|---|
| summary | Our environment contains statistical structure at various scales andalong multiple dimensions. This offers rich opportunity for the brain to exploit this structure for efficient ways to code and process information, e.g., by making and testing predictions about upcoming stimuli. Several theories suggest that making and testing predictions is implemented along cortical processing hierarchies. In this talk, I will address the question how two corecharacteristics of such hierarchies, namely selectivity
(to stimulusfeatures) vis-à-vis invariance, contribute to predictive processing. In the first part of my talk, using the macaque monkey faceprocessing system as a model, I will show that sensory predictions are indeed generated in hierarchical networks of brain areas,and that single neurons in high-level visual cortex can computeprediction errors. Importantly, the hierarchical structure ofvisual processing pathways that gives rise to invariantrepresentations allows constructing abstract predictions not bound to the specific low-level features of learned stimuli.This equips high-level visual cortex with a machinery for generalization that may add additional resources towards the computation of socially highly relevant dimensions through an efficient neural code. While it can be advantageous to abstract from low-level features in some situations and tasks, other conditions may necessitate mechanisms that allow separation of stimulus dimensions for predictive processing. In the second part of my talk, I will present data from intracortical recordings at the laminar scale from human epilepsy patients. Focusing at cross-modal predictions, I will show how predictions about distinct features of stimuli, i.e., their identity and their timing, are implemented in distinct circuits. This factorization of “what” and “when” processing might endow the brain with the flexibility to combine and segregate different kinds of predictability in the dynamic modulation of sensory processing. Together, these studies provide a link between predictive processing and core organizational principles of cortical processing hierarchies. |
|---|
| responsibles | Blancho |
|---|
Workflow history| from state (1) | to state | comment | date |
|---|
| submitted | published | | 2022/09/14 14:26 UTC |
| |
|