Self-organizing spiking neural model for learning fault-tolerant spatio-motor transformations

Area of Science:

• Computational Neuroscience
• Robotics
• Machine Learning

Background:

• Spiking neural networks (SNNs) are biologically plausible computational models.
• Learning spatio-motor transformations is crucial for robotic control and understanding biological movement.
• Existing models often struggle with incomplete or noisy input data.

Purpose of the Study:

• To develop and evaluate a novel spiking neural network model for learning spatio-motor transformations.
• To investigate the model's ability to learn forward and inverse kinematics in a robotic reaching task.
• To assess the model's tolerance to partial sensory or motor input loss.

Main Methods:

• A multilayered spiking neural network architecture using integrate-and-fire neurons.
• Spike-timing-dependent plasticity (STDP) learning rule for synaptic plasticity.
• A 2-degree-of-freedom robot-based reaching task simulating nonlinear function learning.

Main Results:

• The model successfully learned forward and inverse kinematics for the robot reaching task.
• Demonstrated capability in learning complex spatio-motor transformations.
• Exhibited significant tolerance to partial absence of sensory or motor inputs during learning.

Conclusions:

• The proposed spiking neural model effectively learns spatio-motor transformations.
• The model's robustness to input perturbations suggests potential for real-world applications.
• This work provides a foundation for developing more sophisticated learning systems in robotics and neuroscience.