Project

Back to overview

Learning via top-down signaling in the neocortex

English title Learning via top-down signaling in the neocortex
Applicant Senn Walter
Number 118084
Funding scheme Interdisciplinary projects
Research institution Institut für Physiologie Medizinische Fakultät Universität Bern
Institution of higher education University of Berne - BE
Main discipline Neurophysiology and Brain Research
Start/End 01.10.2007 - 30.04.2011
Approved amount 350'000.00
Show all

All Disciplines (3)

Discipline
Neurophysiology and Brain Research
Neurology, Psychiatry
Information Technology

Keywords (11)

learning; concept formation; categorization; synaptic plasticity; top-down bottom-up interactions; computational neuroscience; perceptual learning; visual processing; sensory representation; predictive coding; neuronal implementation

Lay Summary (English)

Lead
Lay summary
The project investigates how learning in the brain can be improved by internal feedback signaling. As an example we consider the neuronal representation of visual stimuli across different cortical areas. Perceptual learning refers to improved sensory detection capabilities based on repeated stimulus presentations. Typically, this form of learning is explained as a refinement of the sensory representation or of its readout. In contrast to this view, we consider perceptual learning as a transient modulation of the early sensory representation through top-down signaling from higher cortical areas. This top-down model is tested against in vitro recordings in monkey visual cortex and psychophysical data from human subjects. We also ask how the cortical representation of visual images must develop in order to minimize putative reconstruction errors based on the neuronal activity. Within this formal framework we show that the representation should naturally split into novelty and familiarity neurons. The response properties of these neurons explain the recordings from visual cortex of monkeys presented with noisy novel and familiar stimuli. The obtained novelty-familiarity representation is also compatible with a predictive coding scenario in which novel information is filtered out at each level of the cortical hierarchy while familiar information is passed through to higher cortical stages. The project offers a theoretical framework for understanding the hierarchy of cortical stimulus representation through feedforward and feedback signaling.
Direct link to Lay Summary Last update: 21.02.2013

Responsible applicant and co-applicants

Employees

Publications

Publication
Sequence learning with hidden units in spiking neural networks
BreaJohanni (2011), Sequence learning with hidden units in spiking neural networks, in Advances in Neural Information Processing Systems 24 (NIPS 2011), 1-9.
Spike-Based Reinforcement Learning in Continuous State and Action Space: When Policy Gradient Methods Fail
Vasilaki Eleni, Frémaux Nicolas, Urbanczik Robert, Senn Walter, Gerstner Wulfram (2009), Spike-Based Reinforcement Learning in Continuous State and Action Space: When Policy Gradient Methods Fail, in PLoS Computational Biology, 5(12), e1000586-e1000586.
Stimulus sampling as an exploration mechanism for fast reinforcement learning
Vladimirskiy Boris B., Vasilaki Eleni, Urbanczik Robert, Senn Walter (2009), Stimulus sampling as an exploration mechanism for fast reinforcement learning, in Biological Cybernetics, 100(4), 319-330.
Dopamine Increases the Gain of the Input-Output Response of Rat Prefrontal Pyramidal Neurons
Thurley Kay, Senn Walter, Lüscher Hans-Rudolf (2008), Dopamine Increases the Gain of the Input-Output Response of Rat Prefrontal Pyramidal Neurons, in Journal of Neurophysiology, 99(6), 2985-2997.
Modulating the granularity of category formation by global cortical states
Kim Yihwa (2008), Modulating the granularity of category formation by global cortical states, in Frontiers in Computational Neuroscience, 2, 1-14.
Perceptual Learning via Modification of Cortical Top-Down Signals
Schäfer Roland, Vasilaki Eleni, Senn Walter (2007), Perceptual Learning via Modification of Cortical Top-Down Signals, in PLoS Computational Biology, 3(8), e165-e165.

Associated projects

Number Title Start Funding scheme
121308 High resolution mass spectrometry for in vivo characterization of bioactive compounds for brain and tumor research 01.10.2008 R'EQUIP
105966 Top-down gain modulation and learning across cortical areas 01.10.2004 Project funding (Div. I-III)
133094 Dendritic pointers and time multiplexing as cortical binding mechanisms 01.05.2011 Project funding (Div. I-III)
133094 Dendritic pointers and time multiplexing as cortical binding mechanisms 01.05.2011 Project funding (Div. I-III)

-