Prediction in cultured cortical neural networks

Area of Science:

• Neuroscience
• Computational Neuroscience

Background:

• Neural networks are theorized to predict input, a function potentially underpinning information processing, motor control, and decision-making.
• While retinal cells and some brain regions show predictive capabilities, it's unproven if this is a universal feature of neural networks.

Purpose of the Study:

• To investigate if random in vitro neuronal networks possess predictive abilities.
• To explore the relationship between prediction and short-term/long-term memory in these networks.

Main Methods:

• Applied focal electrical and global optogenetic stimulation to in vitro neuronal networks.
• Quantified prediction and short-term memory using mutual information.
• Assessed changes over 20 hours of focal stimulation.

Main Results:

• Cortical neural networks demonstrated prediction of future stimuli, primarily driven by immediate network responses.
• Prediction was significantly dependent on short-term memory for both stimulation types.
• Prediction required less short-term memory under focal stimulation.
• This dependency decreased as long-term connectivity changes occurred during prolonged focal stimulation.

Conclusions:

• Random neuronal networks exhibit a generic ability to predict stimuli.
• Short-term memory is crucial for prediction, but its necessity diminishes as long-term memory traces form.
• Long-term memory formation may contribute to efficient prediction in neural networks.