Deep Learning

Objectives

This course is devoted to neural network (NN) architectures and their extension known as deep learning. Beforehand, the stochastic gradient descent algorithm and the back-propagation – its application to feedforward neural networks – are introduced to be further used as the learning basis. This is followed by the study of most spread NN architectures for regression and classification. Among those, convolutional neural networks (CNN) are investigated in detail and other structures like Restricted Boltzmann machines (RBM) and the contrastive divergence algorithm (CD-k) are examined. Further practical aspects will be addressed about the usage of Deep Learning to resolve typical problems like pattern recognition or object detection/tracking. Presented material shall be motivated by the theoretical background together with real data illustrations. There will be specific labs for each topic held in R and Python.

Course outline

– Introduction to deep learning.
– Neural network architectures.
– Stochastic gradient descent and the back-propagation algorithm.
– Neural networks for regression and classification.
– Convolutional neural networks, Restricted Boltzman Machines.
– Applications: Pattern recognition, object detection

Prerequisites

Regression analysis, gradient descent, (matrix) algebra, R, Python (basics).