Supervised learning in multilayer spiking neural networks.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Spiking neural networks (SNNs) offer a biologically plausible model for computation.
• Existing supervised learning algorithms for SNNs face limitations with multi-spike neurons and hidden layers.
• Efficient training of complex SNN architectures is crucial for advancing AI and neuroscience research.

Purpose of the Study:

• To introduce a novel supervised learning algorithm for multilayer spiking neural networks.
• To address the limitations of current algorithms in handling multi-spike neurons and hidden layers.
• To demonstrate the algorithm's versatility across various neuron models and coding schemes.

Main Methods:

• A supervised learning algorithm designed for multilayer spiking neural networks.
• Application to neurons capable of firing multiple spikes within artificial neural networks.
• Compatibility with linearizable neuron models and diverse spike train coding schemes.

Main Results:

• Successfully applied to linearly nonseparable problems like XOR and the Iris dataset.
• Demonstrated effectiveness on complex classification and mapping tasks.
• Validated performance in noisy conditions, showing smaller network requirements than reservoir computing and faster convergence than SpikeProp.

Conclusions:

• The proposed algorithm enables effective supervised learning in complex multilayer SNNs with multi-spike neurons.
• It offers a versatile and efficient approach for training SNNs on challenging benchmark and real-world problems.
• The algorithm represents a significant advancement for SNNs in machine learning and computational neuroscience.