DD2437NeuralNetwork

There are 4 lab sessions.

Lab 1 : Learning and generalisation in feed-forward networks — from perceptron learning to backprop

Main objectives :

• to design and apply networks in classification, function approximation, and generalization tasks
• to identify key limitations of single-layer networks
• to configure and monitor the behavior of learning algorithms for single- and multi-layer perceptrons networks
• to recognize risks associated with backpropagation and minimize them for robust learning of multi-layer perceptrons.

Lab 2 : Radial basis functions, competitive learning and self-organisation

Main objectives :
In this lab, we have used an RBF network to approximate one- and two-dimensional functions. And we have developed a competitive learning algorithm to automate the process of RBF unit initialization. Furthermore, We have implemented the core algorithm of SOM and used it for three different tasks.

Lab 3 : Hopfield Networks

Main objectives :

• Understand the principles underlying the operation and functionality of auto-associative networks
• Train the Hopfield network
• Study the attractor dynamics of Hopfield networks the concept of the energy function
• Understand how auto-associative networks can do pattern completion and noise reduction
• Investigate the question of storage capacity and explain features that help increase it in associative memories

Lab 4 : Restricted Boltzmann Machines and Deep Belief Nets

Main objectives :

• Understand the learning process of RBMs
• Apply basic algorithms for unsupervised greedy pretraining of RBM layers and supervised greedy pretraining of DBN
• Design multi-layer neural network architectures based on RBM layers for classification problems
• Study the functionality of DBNs including generative aspects