Neural Networks

(3 Credit Hours) · Undergraduate

Neural networks are computer models inspired by our understanding of how human brain learns and processes an acquired information. This introductory course will guide you through different neural architectures suitable for various applications. We will start with bio-inspired modeling of neurons (inputs, activation function) and discuss how to use them to build artificial networks (connections, layers). The course will present structures used in supervised learning, that is static networks (Rosenblatt perceptron, Adaline, multi-layer perceptron, radial networks, deep networks including recently popular convolutional nets) and dynamic networks (recurrent networks, associative memory, Hopfield's and Boltzman's machines). Next the networks used in unsupervised learning will be discussed, such as self-organizing maps, Kohonen's networks and structures based on adaptive resonance theory (ART). The course will show how to efficiently apply different types of artificial neural networks in approximation and classification tasks and dynamic systems. We will discuss appropriate learning strategies for each of these applications and their details, such as gradient estimation and minimization techniques. Semester projects will be focused on solving real classification tasks related to computer vision area and with the use of up-to-date neural network software.