Online supervised learning of temporal patterns in biological neural networks under feedback control

Area of Science:

• Neuroscience
• Computational Neuroscience
• Bioengineering

Background:

• In vitro biological neural networks (BNNs) are valuable for studying cell-environment interactions and generating temporal outputs.
• Understanding neural dynamics is crucial for motor control and developing advanced computing paradigms.

Purpose of the Study:

• To develop a real-time closed-loop BNN system for generating diverse temporal signals.
• To investigate the role of feedback and microfluidic control in shaping neural network dynamics.
• To explore BNNs as a platform for understanding cortical computation and neuromorphic computing.

Main Methods:

• Integration of cultured cortical neurons with microfluidic devices and high-density microelectrode arrays.
• Development of a real-time closed-loop system with a linear decoder and feedback.
• Utilizing microfluidic devices for top-down control of network formation.

Main Results:

• The BNN system successfully generated periodic and chaotic temporal signals.
• Training with a linear decoder enabled autonomous learning and generation of diverse temporal patterns.
• Feedback transformed irregular BNN activity into structured, low-dimensional dynamics with stable state transitions.
• BNNs demonstrated adaptability by sustaining oscillations at various target frequencies (4–30 s).
• Microfluidic control suppressed excessive synchronization and increased dynamic complexity, enhancing training and output robustness.

Conclusions:

• The developed closed-loop BNN system provides a biologically inspired platform for studying neural computations.
• This system facilitates the generation of coherent temporal outputs, relevant for motor control.
• The findings advance energy-efficient neuromorphic computing paradigms by leveraging biological neural networks.