Coherent Infomax as a computational goal for neural systems

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Cerebral cortex signal processing is theorized to use a common, multi-purpose algorithm within canonical cortical micro-circuits.
• This algorithm generates distributed, coherent activity patterns, suggesting a unified computational principle across brain regions.

Purpose of the Study:

• To formally specify the objectives and dynamics of the Coherent Infomax theory for cortical processing.
• To enhance the biological relevance of Coherent Infomax by focusing on contextual guidance and Bayesian interpretation.

Main Methods:

• Formal derivation of processing dynamics and learning rules for neural networks.
• Integration of Bayesian principles for contextual guidance of learning and processing.
• Specification of on-line learning rules and computationally feasible approximations.

Main Results:

• Coherent Infomax theory formally specifies objectives and learning rules for neural processing.
• The theory supports a model where local processors use driving and contextual synaptic connections for flexible coding.
• Consistency established between Coherent Infomax and Bayesian interpretations of contextual processing and learning.

Conclusions:

• Coherent Infomax provides a framework for understanding signal processing and learning in the cerebral cortex.
• Contextual modulation is crucial for flexible information processing and adaptation in neural systems.
• The theory offers computationally feasible learning rules for large-scale neural networks.