Emergence of input selective recurrent dynamics via information transfer maximization

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Evolutionary Computing

Background:

• Brain network structures exhibit specialized wiring patterns crucial for function, influenced by genetics and evolution.
• The organizational principles underlying these efficient wiring patterns for information processing remain incompletely understood.
• Understanding these principles is key to deciphering biological information processing and advancing artificial systems.

Purpose of the Study:

• To investigate the underlying principles of neural circuit formation and information processing.
• To frame the optimization of information transmission and maintenance costs as a multi-objective challenge.
• To explore wiring patterns in the visual system using computational models.

Main Methods:

• Utilized information theory to quantify information transmission.
• Employed evolutionary computing algorithms to explore multi-objective optimization.
• Focused on the visual system as a model for circuit formation and information processing.

Main Results:

• Efficient information transmission requires sparse circuits with modular structures and distinct wiring patterns.
• Significant trade-offs exist, emphasizing the need for balance in wiring pattern development.
• Effective circuits demonstrate moderate flexibility in response to stimuli, aligning with visual system studies.

Conclusions:

• Maximizing information transfer can lead to self-organization of information processing functions, similar to biological circuits.
• These findings offer insights into neuroscience and the potential for improving reservoir computing.
• The study highlights the importance of balancing efficiency and cost in neural network design.