attern Recognition (PR) is the scientific discipline that studies the operation and design of systems that recognize patterns in data (static or dynamic). The course will give the algorithms able to explore many of the PR applications, and the general analytical models able to cope with real world data. We will be focused on the basic concepts, the models and the tools necessary to the comprehension and design of a pattern recognition system. Starting with a conceptual discussion involving the nature and difficulties inherent to the PR problem, we will give the notions of pattern discrimination, decision functions and decision regions, class separability and metrics. Then we will proceed with the study of the feature extraction and feature selection, parametric and non-parametric methods, dimensionality reduction and kernel methods. We will end with the design and assessment of classifiers for pattern recognition.
1. Pattern Discrimination: decision functions and decision regions; class separability metrics; Linear Discriminants (Euclidian and Mahalanobis), and Fisher discriminant.
2. Feature extraction and feature selection; feature ranking; Kruskal Wallis. Data pre-processing (outliers removal, normalization and scaling, missing data)
3. Clustering: Hierarchical and k-means algorithms
4. Parametric Methods: model selection, linear generalized models, mixture models, Bayes Classification, Parameter estimation: likelihood method; Bayes and risk estimation; Maxima A Posteriori (MAP); classifier; Kullback-Leibler divergence
5. Non-parametric methods: density estimation: Parzen windows and K-nearest neighbors.
6. Dimensionality reduction; Principal Component Analysis (PCA); Non-linear methods.
7. Kernel methods: Mercer kernel, Kernel PCA.
8. Classifier assessment sampling, confusion matrix and error probability; ROC curves; Bootstrapping, Boosting
. Competence in analysis and synthesis; . Computer Skills for the scope of the study; . Competence to solve problems; . Capacity of decision; . Critical thinking; . Competence in oral and written communication; . Adaptability to new situations; . Creativity; . Competence in applying theoretical knowledge in practice; . Research skills; (by decreasing order of importance)
Teaching hours per semester
total of teaching hours
assessment implementation in 20172018 Assessment Report of a seminar or field trip: 25.0% Project: 35.0% Exam: 40.0%
Bibliography of reference
1. Bishop, C.M., ?Pattern Recognition and Machine Learning?, Springer Verlag, 2006
2. Duda, R. O., Hart, P.E., and Stork, D.G., ?Pattern Classification,? 2nd ed. Wiley Interscience (2001)
3. J.P. Marques de Sá, ?Pattern Recognition: Concepts, Methods and Applications?, 2001, XIX, 318 p., 197 illus., Springer-Verlag (2001)
4. M. N. Murty and V. S. Devi, ?Pattern Recognition: An Algorithmic Approach?, Springer, 1st Edition., XII, 263 p. (2011)
Theoretical classes with detailed presentation, using audiovisual means, of the concepts, principles and fundamental theories and solving of basic practical exercises to illustrate the practical interest of the subject and exemplify its application to real cases.
Theoretical-practical classes where the students solve practical exercises, which require the combination of different theoretical concepts and promote critical reasoning in the presence of more complex problems.