Title of Submission
Electrical and Computer Engineering, PhD
Major Advisor Name
Type of Submission
Deep learning, i.e., the use of deep convolutional neural networks (DCNN), is a powerful tool for pattern recognition (image classification) and natural language (speech) processing. Deep convolutional networks use multiple convoltuion layers to learn the input data. They have been used to classify the large dataset Imagenet with an accuracy of 96.6%. In this work deep spiking networks are considered. This is new paradigm for implementing artificial neural networks using mechanisms that incorporate spike-timing dependent plasticity which is a learning algorithm discovered by neuroscientists. Advances in deep learning has opened up multitude of new avenues that once were limited to science fiction. The promise of spiking networks is that they are less computationally intensive and much more energy efficient as the spiking algorithms can be implemented on a neuromorphic chip such as Intel’s LOIHI chip (operates at low power because it runs asynchronously using spikes). Our work is based on the work of Masquelier and Thorpe, and Kheradpisheh et al. In particular a study is done of how such networks classify MNIST image data and N-MNIST spiking data. The networks used in consist of multiple convolution/pooling layers of spiking neurons trained using spike timing dependent plasticity (STDP) and a final classification layer done using a support vector machine (SVM).