Research Paper
Optimizing CSNN (Convolutional Spiking Neural Networks) Model for
Surface Textures detection
Research Paper
My research work in Computational Neuroscience involves investigating the effect of integrating Spike timing dependent plasticity learning method and population encoding on Convolutional Spiking Neural Network (CSNN) on the time taken for the algorithm to converge.
The CSNN referred in the paper is being developed to predict surface textures. It harnesses data optimization techniques of CNN and extends it to generate spike train for each class of surface textures. The given input neuron spike when processed through trained CSNN model, generates distinct output population code pattern for each texture class, aligned with biological neural activity.
​
A natural extension to this work, is to train and optimize CSNN model to identify skin lesion textures into different Melanoma subtypes.
Abstract
Optimization techniques to improve
Convolutional Spiking Neural Networks Model for Surface Textures detection
Spiking Neural Networks (SNNs) are a third generation type of artificial neural network that are far more energy efficient, faster, biologically plausible, and dynamic. However, they are quite difficult to train due to the lack of a generally established learning method for them.
​
Population Coding, a type of neural coding, maps naturalistic stimuli to spike trains and vice versa. Such a biological mechanism could be integrated into Convolutional Spiking Neural Network architecture for processing spatio-temporal information.