Neural Networks and Genetic Algorithms

The course will be articulated in three days, comprising 3, 3 and 4 hours of class lectures, respectively. More details can be found in the attached sheet.
  1. Introduction to artificial neural networks. Main characteristics of neural networks. Artificial neuron model. Activation functions. Types of neural networks. Supervised learning. Delta rule. Convergence of the delta rule. Learning rate. Local minima. Batch and online learning. Momentum. Perceptron. Perceptron learning rule. The XOR problem. Hidden layers. Multilayer networks. Multilayer perceptron. Error backpropagation. The backpropagation algorithm. (3h)
  2. Creation of the training set: analysis of the data, selection of variables, data preprocessing, outliers, missing data, non-numeric data, data normalization. Stopping conditions of neural network training. Overtraining. Early stopping. Train, test and validation sets. Network's size. Unbalanced data sets. Examples of applications. Radial functions. Introduction to RBF networks. Learning strategies for RBF networks. Comparison of RBF networks and Multilayer Perceptrons. Unsupervised learning. Competitive networks. Self-organizing maps. Examples of applications. (3h)
  3. Introduction to deep learning. Convolutional neural networks. (2h)
  4. Introduction to Genetic Algorithms. Crossover and mutation. Fitness function. Selection for recombination. Selection for replacement and survival. Binary encoding. Real-valued encoding. Examples of applications. (2h)
Giovanni Mengali,
30 mag 2018, 02:46