87994 - PATTERN RECOGNITION

Course Unit Page

Academic Year 2018/2019

Learning outcomes

At the end of this course the student will learn the principles and commonly used paradigms and techniques of pattern recognition. In particular he/she will be able to: - explain and compare a variety of pattern classification techniques, - apply performance evaluation methods for pattern recognition, - apply pattern recognition techniques to real-world problems in the Applied Physics field, - implement simple pattern classifiers, - demonstrate successful applications to process and analyze data (e.g. images), and make automatic decisions based on extracted feature information.

Course contents

24 hours (3 CFU) Castellani

Introduction Probability Theory, Probability densities, Expectations and covariances, Bayesian probabilities, Bayesian curve fitting, Model Selection, the Curse of Dimensionality, Statistical Inference and decision, loss functions for regression

Probability Distributions Binary and Multinomial Variables, Beta, Dirichlet and Gaussian Distribution, Gaussian Mixtures, the Exponential Family, maximum likelihood and sufficient statistics, Conjugate priors, Noninformative priors, Nonparametric Methods Inference and association test

Linear models for Regression and Classification Linear Basis Function Models, Bias-Variance Decomposition, Bayesian Linear Regression, Bayesian Model Comparison, Discriminant Functions, Probabilistic Generative and Discriminative Models

Neural Networks Learning rules, Hebbian,BCM and Hopfield model, Feed-forward Network Functions, Network Training, Error Backpropagation, Regularization,

Introduction to Kernel methods and Graphical models

24 hours (3 CFU) Remondini

Feature extraction & optimal projection:

Principal Component Analysis, Singular Value Decomposition, Factor Analysis, Multi-Dimensional Scaling, ISOMAP, Relation to cost function maximization and eigenvalue problem (Rayleigh quotient), Practical aspects: data regularization and normalization.

Supervised classification:

Support Vector Machine, Discriminant Analysis, Random Forest.

Unsupervised clustering:

Hierarchical, k-means, spectral clustering techniques.

Introduction to Deep Learning:

Error minimization: the backpropagation algorithm.,Feedforward Neural Networks., Convolutional Neural Networks.

Robust analysis

Bias-variance dilemma, Best practices, Crossvalidation procedures

Suggested books:

Readings/Bibliography

Bishop - Machine learning and pattern recognition

Tibshirani Tusher - Methods of statistical learning

Teaching methods

Slides and blackboard

Assessment methods

Project, literature study and questions.

Teaching tools

Programming environment and server connection.

Office hours

See the website of Gastone Castellani

See the website of Daniel Remondini