| TSBB41 | Classification, Learning and Neural Nets, 6 ECTS-points /Klassificiering, inlärning och neuronnät/ Advancement level: D | |
| Aim: The course introduces methods and computing structures for learning and self-organization. The course describes how such methods can be used to find meaningful relations in multidimensional signals where the degree of complexity makes traditional model-based methods unsuitable or impossible to use. In practice such signals come as a rule rather than an exception. Examples of application areas are function approximation, pattern recognition, content addressable memories, prediction, optimization, process control and classification. Many methods, but not all, have been developed inspired by the function of the brain and a general aspiration to develop computing structures having features such as adaptation, ability to learn, fault tolerance, ability to generalization and extrapolate, distributed knowledge representation and massive parallelism. Examples of areas where techniques based on learning has proven to be competitive are industrial process optimization (paper/pulp, steel, ore), economical prediction of markets, text and speech recognition, document searching and image- and image-sequence analysis. Prerequisites: General: Linear algebra, Calculus, Mathematical statistics, Control theory, Signal theory. Spec: Multi-dimensional signal analysis TSBB 30, Computer Vision TSBB 02. Course organization: Lectures, Exercises, Laboratory exercises (incl obligatory home work) Course content: Classification: - Pattern Recognition - Discriminant functions - Statistical methods - Clustering Content addressable memories: - State spaces - Hopfield memories - Auto- and Hetero-associative memories Supervised learning: - The perceptron - The multi-layer perceptron - Stochastic gradient search - The "Error back-propagation" algorithm Unsupervised learning: - Principal component analysis (PCA) - "Winner take all" algorithms - Topology preserving methods - Self organizing maps (SOM) Reinforcement learning: - Markov models - Reward/Punishment methods - Temporal difference methods (TD) - Q-learning Genetic methods: - Genetic algorithms - The two-armed bandit - Genes and schemas - Genetic programming Course literature: D.H. Ballard: "An Introduction to Natural Computation", MIT press 1997 B. A. Kröse, P. van der Smagt: "An Introduction to Neural Networks", kurskompendium Exempelsamling `Kompletterande material' Lab-PM | ||
| TEN1 | Written examination, 2,5 p. | |
| LAB1 | Laboratory work., 1,5 p. | |