Anisotropic connectivity implements motion-based prediction in a spiking neural network

Area of Science:

• Computational Neuroscience
• Systems Neuroscience

Background:

• Predictive coding theory suggests the brain actively infers sensory input for world representation.
• The precise implementation and functional role of connectivity in spiking neural networks for predictive coding remain unclear.

Purpose of the Study:

• To model predictive coding in spiking neural networks using biologically inspired connectivity.
• To investigate the role of connection delays and tuning selectivity in implementing predictive coding.
• To assess the network's ability to predict sensory trajectories.

Main Methods:

• Developed a conductance-based integrate-and-fire neuron network model.
• Modeled retinotopic cortical area architecture with specific connection delays and tuning selectivities.
• Tested the network's predictive capabilities using a moving dot experiment.
• Compared performance against networks with random or isotropic connectivity.

Main Results:

• The proposed network model successfully demonstrated motion-based prediction.
• Networks with random or isotropic connectivity failed to predict the moving dot's trajectory when it disappeared.
• Neuronal spiking activity could be decoded into probabilistic trajectory estimates using linear methods.

Conclusions:

• Network connectivity, specifically connection delays and tuning properties, is crucial for implementing predictive coding in spiking neural networks.
• Biologically inspired network architectures facilitate accurate sensory prediction.
• Simple decoding methods can effectively extract predictive information from neural activity.