Intelligent Systems
The course assumes that students have a background acquired on a Bachelor's Degree in Computer Science. Specifically, Students will be expected to be familiar with Vector and Matrix Calculation and Statistics.
Course Objectives and Expected Outcomes
The course provides a broad coverage of intelligent systems solving pattern recognition problems. Theoretical concepts in intelligent systems and techniques relevant in real-life applications will be examined. At the end of the course, students will have the ability to appropriately choose the proper technique to solve problems of recognition and /or automatic classification of multidimensional data in several domains.
Upon completion of this course, the students will be able to:
• Examine the prerequisites of a classification method
• Apply feature extraction and selection techniques
• Use statistical techniques with expertise in the limits and strengths of each approach
• Investigate the representative power of supervised neural models, apply heuristics in the choice of network topology, check the condition of overfitting, evaluate performances in training and test.
• Compute Flat Clustering, develop sensitivity analysis and parameter setting
• Investigate the potentialities of Competitive and Self-Organizing approaches
• Design a Fuzzy Set-based System, generate IF-Then Rules
• Evaluate performances of an Intelligent System
• Describe main characteristics of Intelligent Systems for Image Classification, Text Categorization, Multidimensional Biomedical Data Analysis
Evaluation procedure
The evaluation procedure consists of a written examination lasting 2 hours. The test consists of 5 questions in general, some entirely of theory, some including numerical exercises. After correction, the student is invited for a review of the written test.
• Introduction to Artificial Intelligence and Pattern Recognition: Historical Perspective, State of the Art of methods and applications
• Decision-theoretical approaches; supervised classification; learning by example: principles; multidimensional patterns
• Feature extraction and selection: Principal Component Analysis,, Pearson correlation coefficient, Information Gain, Sequential Forward Selection
• Multidimensional Patterns in Text Classification, Image Recognition and in Content-Based Retrieval
• Minimum distance classifier
• Bayes’s rule
• Maximum likelihood classifier
• K-Nearest Neighborhood classifier
• Neural Networks: Introduction
• Recurrent Neural Networks: Bidirectional Associative networks, Hopfield Networks
• Feed forward Models:, Perceptron
• Fixed increment rule, Delta rule
• Limits of the perceptron model: XOR problem
• Multilayer Perceptron, topology, back-propagation learning rule
• Applications
• Clustering: principles
• K-means,
• Agglomerative Hierarchical Clustering: Single linkage, Complete linkage
• Competitive Neural Learning
• Self Organizing Maps; Kohonen Networks
• Soft Computing: Fuzzy Sets Theory, Approximate Reasoning, Neuro-Fuzzy approaches, Fuzzy C-means algorithm
• Evaluation metrics, evaluation indexes: Kappa index, Jaccard, Dice coefficients; Applications of evaluation metrics in several domains
• Design of Intelligent Systems; Demo.
Textbook: R.O. Duda, P.E. Hart, and D.G. Stork, Pattern Classification, New York: John Wiley & Sons, 2001
Additional readings, including lecture notes, slides, selected papers from the literature and link s to on line demo will be posted periodically on the class website.