The course introduces the basic principles of various computational methods of data processing where solutions are learned by mimicking natural processes rather than designed. Sometimes grouped under the umbrella term "evolutionary algorithms", these techniques are closely related, yet clearly distinct, from the ones now filed under the "machine learning" label. Solutions are searched or built through trial-and-error using bottom-up approaches and (meta)heuristics, using mechanisms inspired by natural evolution or animal behaviors. The course puts the different paradigms into a historical perspective; differences and correlations among the different approaches are shown, together with the possible synergies with other paradigms.

* Some programming experience with Python 3 * Suggested: Algorithms and Programming * Not required: mathematics and statistics * Not required: knowledge of specific toolkits (e.g., Jenetics, scikit-learn, Open BEAGLE, PyTorch, Keras/TensorFlow)

* Hill-climbing, simulated annealing, iterated local search, variable neighborhood search * Simple Evolution Strategies: (1+1), (1+¿) and (1,¿) * Unified approach to Evolutionary Algorithms * Parameter optimization (Evolution Strategies, Differential Evolution) * Symbolic regression (Genetic Programming) * Swarm intelligence (Ant Colony Optimization, Particle Swarm Optimization) * Memetic Algorithms (hybridization) * Model fitting (Estimation of Distribution Algorithm) * Multi-objective optimization

On site

Written report presentation

P.D.1-1 - December