The course considers foundational and advanced pattern recognition methods for classification tasks in signals and data. We take a Bayesian approach in this course. Simple example applications can be a digit recognition task, or automatic word recognition task. A complex application can be in medical field, such as recognition of disease from patient data.
After passing the course, students should be able to:
- Describe models for Pattern Recognition system and formulate the optimal cost functions in a probabilistic framework. Then analytically and experimentally estimate recognizer performances.
- Describe a pattern recognition problem for a sequence of observed signals and address the problem using hidden Markov models (HMM).
- Design systems and algorithms for pattern recognition. Critically compare the algorithms in a trade-off between complexity and performance. Present and report the results.
- Implement and analyze machine learning based methods for automatic training of pattern recognition systems.
The course considers foundational and advanced pattern recognition methods for classification tasks in signals and data. We take a Bayesian approach in this course. Simple example applications can be a digit recognition task, or automatic word recognition task. A complex application can be in medical field, such as recognition of disease from patient data. The course covers following.
- Pattern recognition problems in Bayesian framework. Forming optimal cost functions, and then establishing maximum-likelihood (ML) and maximum-a-posteriori (MAP) rules for classification.
- Discriminant functions.
- Hidden Markov models (HMM) for classification of sequence of feature vectors
- Machine learning based HMM training - using expectation-maximization (EM)
- Approximate machine learning, such as variational Bayes.