Desegregation of neuronal predictive processing
Bin Wang1, Nicholas J Audette2, David M Schneider2
1Department of Physics, University of California San Diego, La Jolla, CA, 92093, USA.
Biorxiv : the Preprint Server for Biology
|August 16, 2024
Summary
Neural circuits build internal world-models for behavior. This study reveals distributed, multi-modal predictive representations in recurrent networks, challenging specialized cell-type theories and advancing understanding of brain computation.
Area of Science:
- Computational neuroscience
- Systems neuroscience
- Cognitive neuroscience
Background:
- Neural circuits generate internal 'world-models' to guide behavior.
- The predictive processing framework suggests neural activity signals sensory predictions and computes prediction-errors.
Purpose of the Study:
- Investigate the emergence of high-dimensional, multi-modal predictive representations in recurrent networks.
- Understand how the brain generates predictions for complex sensorimotor signals.
Main Methods:
- Simulated recurrent neural networks with varying excitatory/inhibitory balance.
- Probed predictive-coding circuits experimentally using stimuli designed to violate learned expectations.
Main Results:
- Robust predictive processing emerged in networks with loose excitatory/inhibitory balance.
- Found desegregation of stimulus and prediction-error representations, contrary to specialized cell-type hypotheses.
- Model predictions for the roles of specific neuron types and layers in multi-modal prediction were confirmed experimentally.
Conclusions:
- Neural representations of internal models are highly distributed yet structured for flexible behavioral readout.
- The unified framework advances understanding of internal model computation across species.
- Revealed distinct functional roles for excitatory/inhibitory neurons and laminar hierarchy in multi-modal prediction.
