Deep Learning Course (NYU, Spring 2020) - Week 3 – Lecture: Convolutional neural networks

Course website: http://bit.ly/DLSP20-web Playlist: http://bit.ly/pDL-YouTube Speaker: Yann LeCun Week 3: http://bit.ly/DLSP20-03 0:00:00 – Week 3 – Lecture LECTURE Part A: http://bit.ly/DLSP20-03-1 We first see a visualization of 6-layer neural network. Next we begin to the topic of Convolution and Convolution Neural Networks (CNN). We review several types of parameter transformation in CNN and introduce the idea of a kernel, used to learn features in a hierarchical manner, and to classify our input data is the basic idea of a CNN. 0:00:05 – Visualization of Neural Networks 0:07:57 – Parameter Transformations, the Convolution Operator, and Deep Convolutional Neural Networks 0:37:34 – Inspirations from Biology LECTURE Part B: http://bit.ly/DLSP20-03-2 We give an introduction on CNN evolutions. We discuss in detail on architectures of CNN with modern implementation of LeNet5, exemplified by the task of digit recognition on MNIST. Based on its design principles, we expand on the advantages of CNN which fully explores compositionality, stationarity, and locality features of natural images. 0:49:09 – The First ConvNets 1:03:50 – LeNet5 and digit recognition 1:20:27 – Feature Binding and What are ConvNets Good for?