USC Viterbi School of Engineering
Duration: -
Location: Viterbi School of Engineering, University of Southern California, LA, California, USA
Skills: TensorFlow , PyTorch , Theoritical ML , Neural Networks
As one of the 18 people across India to be selected for IUSSTF-Viterbi Summer Research Program, I spent my summer of 2018 working under the guidance of Prof. Meisam Razaviyayn and did the following: • A literature study of previous works related to Quantization and Binarization of Neural Networks • Came up with the idea of adaptively quantizing different layers of a neural network with different number of bits • Modified existing Tensorflow and Pytorch models to test out various hypothesis to improve accuracy • Studied the effect of different regularizers to aid in the quantization/binarization of the layers of a neural network • Developed an algorithm using regularization to optimally choose the quantization for each layer during training