Lifelike Neuromorphic Learning Networks (LNLN)

Abstract

Artificial Neural Network (ANN) has been known and used extensively to solve the demanding tasks of Machine Learning (ML) and Artificial Intelligence (AI). These networks have proven to be exceedingly successful with challenging tasks but only at the cost of doing massive amounts of computations. Spiking Neural Network (SNN) are known to be able to perform the same tasks but potentially with less power and computations. The proposed research develops an application on Spiking Neural Networks Simulators using various algorithms and input encoding to achieve accuracy that is at par with Analog Artificial Neural Network (AANN). Backpropagation approach is used on a pre-trained neural network and it is converted to SNN for rate coding. To add further Spike-timing-dependent plasticity (STDP) is used for training a rate encoded network. Using the above settings significant accuracy is achieved proving its uniqueness amongst the state-of-the-art algorithms. A detailed profiling of current literature is included. These findings underlie a huge potential and may locate the stage for further thrilling novel advances that drives key applications in neuromorphic engineering.