Spike-Timing-Dependent Plasticity With Activation-Dependent Scaling for Receptive Fields Development.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Artificial Intelligence

Background:

• Spike-timing-dependent plasticity (STDP) is a biologically inspired unsupervised learning rule.
• Developing effective unsupervised learning mechanisms for spiking neural networks (SNNs) is an ongoing challenge.

Purpose of the Study:

• To propose a novel variant of STDP with an activation-dependent scale factor for unsupervised learning in SNNs.
• To demonstrate the efficacy of this STDP variant combined with competitive learning for developing receptive fields (RFs).
• To explore the interplay between synaptic scaling and lateral inhibition in RF development.

Main Methods:

• Implementation of a modified STDP learning rule incorporating an activation-dependent scale factor.
• Integration of the proposed STDP with competitive learning for unsupervised RF development.
• Evaluation of the learning rule on the MNIST dataset using classification tasks.

Main Results:

• The proposed STDP variant combined with competitive learning effectively develops receptive fields.
• High accuracy (94.65% single, 95.17% committee) achieved on MNIST, comparable to state-of-the-art unsupervised SNNs.
• Demonstrated the crucial role of synaptic scaling dynamics in developing neuronal selectivity.
• The training process results in sparse data representation and potential for feature detection.

Conclusions:

• The enhanced STDP rule provides an efficient and simple algorithm for unsupervised learning in SNNs.
• The mechanism is effective for developing receptive fields and achieving high classification accuracy.
• Theoretical analysis guarantees convergence for linear Poisson neurons by conserving total synaptic strength.