Abstract

NB —this week’s talks will take place in the Club Room due to maintenance work in the Wolfson Hall.

During the past few years, significant improvement in the task of image recognition has been made with the help of convolutional neural networks. This talk will describe how neural networks work and how they can be trained using back-propagation and the gradient descent algorithm. We shall also investigate how, by utilizing the notion of replicated features, convolutional NN can substantially improve the accuracy of machine object classification. Finally, we will see how human vision differs from computer vision by discussing the concept of inceptionism and by describing how Deep Dream allows us to visualize what the neural network has learned at each layer.