Periodic Reporting for period 1 - INTEREP (Do cortical feedback connections store statistical knowledge of the environment?)
Période du rapport: 2018-03-01 au 2020-02-29
Schizophrenic patients show difficulties in learning and representing behaviorally relevant percepts when the selection of relevant information is determined by previous experience. In this project, we aimed at determining how our knowledge of world,inferred from past experience is stored in the brain. While work on the influence of internal representations on sensory processing has been conducted from a cognitive or computational point-of-view, this study approaches the question at the brain circuit level.
Vision arises from a set of hierarchically-organized areas where neurons in early cortical stages represent simple attributes (such as orientation or direction of line elements) while areas higher in the hierarchy encode progressively more complex aspects of the world. The buildup of these increasingly complex representations is thought to depend on feedforward connections, those that link lower to higher order areas. However, a large number of feedback (FB) projections reciprocally link higher areas to lower areas. Given the complex attributes they represent, FB projections are ideally positioned to influence the sensory responses of lower-order areas carrying predictions based on the learned statistics of visual inputs. In this project, we investigated whether prior knowledge about the world is stored in the connectional specificity of FB axons.
By investigating the specificity of functional wiring of feedback loops in the brain, this project might also inspire the fields of computer science and information technology. Neuronal network-inspired structures could use the rules of feedback wiring established in this proposal – with wiring instructed by input statistics – to design better performing machine learning algorithms.
Vision arises from a set of hierarchically-organized areas where neurons in early cortical stages represent simple attributes (such as orientation or direction of line elements) while areas higher in the hierarchy encode progressively more complex aspects of the world. The buildup of these increasingly complex representations is thought to depend on feedforward connections, those that link lower to higher order areas. However, a large number of feedback (FB) projections reciprocally link higher areas to lower areas. Given the complex attributes they represent, FB projections are ideally positioned to influence the sensory responses of lower-order areas carrying predictions based on the learned statistics of visual inputs. In this project, we investigated whether prior knowledge about the world is stored in the connectional specificity of FB axons.
By investigating the specificity of functional wiring of feedback loops in the brain, this project might also inspire the fields of computer science and information technology. Neuronal network-inspired structures could use the rules of feedback wiring established in this proposal – with wiring instructed by input statistics – to design better performing machine learning algorithms.
To investigate whether sensory experience plays a role in the connectional specificity of FB axonal wiring in the visual cortex, we raised mice in total darkness after weaning. Two photon calcium imaging of neuronal activity allows for measuring the spatial extent of FB axon targeting onto their downstream neurons. From these measurements, it appears that depriving visual inputs results in a more constrained functional wiring of FB axons. Ongoing works aim to determine if the connectional specificity of FB axons wiring reflects the sensory statistics of the visual world.
Cutting-edge dual-color, volumetric two-photon calcium imaging permitted the measurement of the functional organization or wiring of FB axons relative to their neuronal target. The altered connectional specificity of FB wiring in dark-reared animals proves the necessity of sensory experience in the precise organization of wiring between brain regions. Ongoing experiments will test whether statistical regularities in visual inputs are reflected in the exact organization of visual FB wiring. In future experiments, we aim to manipulate the spatio-temporal dynamics of pre- and post-synaptic activities to assess the mechanistic implementation of such wiring in the brain.
We expect this project to have several repercussions: 1) it will provide a wider understanding of connection organization in the brain, by providing a functional description of connection targeting, 2) it will constrain theories of feedback and argue for or against the role of predictive coding and/or in learning through backpropagation,3) it might inspire new generations of machine learning algorithms inspired from neuronal network architecture, 4) it might pave the way for a better understanding of altered prior utilization seen in schizophrenic patients.
We expect this project to have several repercussions: 1) it will provide a wider understanding of connection organization in the brain, by providing a functional description of connection targeting, 2) it will constrain theories of feedback and argue for or against the role of predictive coding and/or in learning through backpropagation,3) it might inspire new generations of machine learning algorithms inspired from neuronal network architecture, 4) it might pave the way for a better understanding of altered prior utilization seen in schizophrenic patients.