How attention can create synaptic tags for the learning of working memories in sequential tasks

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Intelligence relies on learning appropriate responses to new stimuli.
• Neurons in the association cortex are crucial for intelligence, developing persistent activity during learning.
• The precise mechanisms of this learning process remain poorly understood.

Purpose of the Study:

• To propose a biologically plausible learning scheme explaining how trial-and-error learning induces neuronal selectivity and working memory.
• To elucidate the role of attentional feedback and neuromodulators in synaptic plasticity for learning.
• To demonstrate how neural networks can create new working memory representations.

Main Methods:

• Development of a novel learning scheme based on attentional feedback signals and neuromodulation.
• Modeling how synaptic tags are formed at relevant connections.
• Analysis of how tagged synapses interact with neuromodulators to control plasticity.

Main Results:

• The proposed learning rule enables neural networks to form working memory representations via persistent activity.
• This generic rule explains learning for both linear and non-linear stimulus-response mappings.
• The scheme accounts for tuning to category boundaries, analog variables, and probabilistic evidence integration.

Conclusions:

• The developed learning scheme provides a mechanistic explanation for how association neurons acquire selectivity and working memory capabilities.
• This model offers a unified framework for understanding diverse learning phenomena in the brain.
• The findings have implications for understanding intelligence and developing artificial learning systems.