Fast efficient coding and sensory adaptation in gain-adaptive recurrent networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Sensory Processing

Background:

• Neural sensory systems must adapt to changing environmental statistics to maintain accurate representations.
• Neuronal tuning curves are hypothesized to optimize to the prior stimulus distribution under efficient coding principles.
• Empirical observations of 'adapter repulsion' in tuning curves contrast with theoretical 'prior attraction' predictions, leaving the underlying mechanism unclear.

Purpose of the Study:

• To investigate the mechanisms of rapid neural adaptation in sensory systems.
• To reconcile the contrasting phenomena of adapter repulsion and prior attraction.
• To propose a unified theoretical and mechanistic model for efficient coding in recurrent sensory networks.

Main Methods:

• Development of a gain-adaptive, recurrent sensory network model.
• Incorporation of an efficient coding objective balancing accuracy and spiking cost.
• Analysis of adaptive tuning curve dynamics under varying prior distributions.

Main Results:

• The model demonstrates that modulated gains propagating through the network lead to quickly adaptive tuning curves.
• The model successfully accounts for adapter repulsion under peaked prior distributions.
• The model predicts and is supported by behavioral evidence for fast prior attraction under broader distributions.

Conclusions:

• Gain modulation in recurrent circuits provides a unified mechanism for rapid adaptation in neural sensory systems.
• The proposed framework reconciles seemingly contradictory adaptive phenomena observed in neural tuning curves.
• This work advances our understanding of efficient coding principles and their implementation in biological neural networks.