Related Experiment Videos
Retrospective and prospective persistent activity induced by Hebbian learning in a recurrent cortical network
Gianluigi Mongillo1, Daniel J Amit, Nicolas Brunel
1Dipartimento di Fisiologia Umana, Università di Roma La Sapienza, Rome, Italy INFM, Dipartimento di Fisica, Università di Roma La Sapienza, Rome, Italy.
The European Journal of Neuroscience
|November 19, 2003
Summary
This study models how the brain learns long-term associations using persistent neuronal activity in visual working memory. The model shows how Hebbian learning strengthens connections, leading to predictive neural activity for associated stimuli.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Cognitive Science
Background:
- Persistent neuronal activity in the associative cortex is linked to visual working memory and associative memory formation.
- Delayed pair-associate tasks reveal neural correlates of long-term memory for stimulus associations.
Purpose of the Study:
- To model a recurrent cortical network with Hebbian plastic synapses to understand the learning of long-term associations.
- To investigate the mechanisms underlying retrospective and prospective activity during associative learning.
Main Methods:
- A recurrent cortical network model with Hebbian plastic synapses was used.
- The model was subjected to a pair-associate learning protocol mimicking experimental tasks.
- The influence of synaptic properties, like N-methyl-d-aspartate receptor fractions, on activity transitions was analyzed.
Main Results:
- Learning initially produced retrospective delay activity (representing individual images).
- With further learning, prospective activity emerged, activating associated stimulus representations before presentation.
- The model demonstrated that connection strength, dependent on training frequency, governs prospective activity, and synapse dynamics influence transition speed.
Conclusions:
- Persistent activity in working memory can drive the learning of long-term associations.
- The model successfully reproduces neuro-physiological data and offers testable predictions.
- Synaptic plasticity mechanisms, modulated by receptor dynamics, are crucial for associative learning and predictive neural representations.