|View general information Description Information prior to enrolment Learning objectives and results Content View the UOC learning resources used in the subject Additional information on support tools and learning resources Guidelines on assessment at the UOC View the assessment model|
|This is the course plan for the second semester of the academic year 2023/2024. To check whether the course is being run this semester, go to the Virtual Campus section More UOC / The University / Programmes of study section on Campus. Once teaching starts, you'll be able to find it in the classroom. The course plan may be subject to change.|
The main motivation of this subject is to make known the automatic learning and how it is placed within the field of the Artificial intelligence.
In Artificial Intelligence (Degree in Computer Engineering) an overview of the Artificial Intelligence was given and some of the so-called classical methods and techniques, such as problem solving and search and knowledge-based systems, more advanced techniques such as neural networks and approximate reasoning were also seen. In the subject of Learning Computational problems for learning (supervised and unsupervised) and multi agent systems were introduced. In this subject you will deepen into advanced learning problems, introducing features extraction systems, non-linear systems based on kernels, optimization processes or deep learning techniques, always from a practical side touching examples of real problems.
It is advisable to have studied the subjects of Artificial Intelligence and Computational Learning of the degree in Computer Engineering.
It is also highly recommended to have passed the subject of programming practices or equivalent in some computer science program.
Although the course is not designed to have a high programming load, the most basic concepts of algorithmic will be taken for granted.
The general competences of the Master in this subject are:
The specific competences of this subject are:
In this subject the contents have been structured in two modules.
The first module shows an overview of learning within Artificial Intelligence. In principle, the distinction between algorithms dedicated to clustering and recommendation of information, algorithms of extraction and selection of characteristics, classification algorithms, optimization techniques and deep learning techniques. The distinction between supervised and unsupervised learning is inherent in the clustering and classification chapters, even though it is also present in the feature extraction chapter.
The second module, much shorter, is dedicated to learning the Python language. It is intended to introduce the student to some (of the many) characteristics that this language has, with a view to a better understanding of the modules of theory, and to be able to practice independently.
The detailed content of each of these modules is given below.
1. Introduction to Artificial Intelligence (AI)
1.1. Neurons and Transistors
1.2. Brief history of AI.
1.3. Areas of application
2. Recommenders and clustering
2.1. Metrics and measures of similarity
2.1.1. Application example
2.1.2. Euclidean distance
2.1.3. Pearson's Correlation
2.2. Recomender systems
2.2.1. General concepts
2.2.2. Surprise library
2.2.3. Nearest Neighbours
2.2.4. Singular Value Decomposition
2.3. Clustering algorithms
2.3.1. Application example
2.3.2. General concepts
2.3.3. Hierarchical clustering. Dendrograms.
2.3.5. Fuzzy K-means
2.3.6. Spectral Clustering
2.3.7. Model Based Recommenders
3. Extraction and selection of attributes
3.1. Matrix factorization techniques.
3.1.1. Singular value decomposition ¿¿(SVD)
3.1.2. Principal Component Analysis (PCA)
3.1.3. Independent Component Analysis (ICA)
3.1.4. Non-negative matrices factorization (NMF)
3.2. Data Discrimination in Classes
3.2.1. Linear discriminant analysis (LDA)
3.3. Visualization of multidimensional data
3.3.1. Multidimensional scaling (MDS)
4.1.1. Categorization of texts
4.1.2. Automatic classification learning
4.1.3. Typology of algorithms by classification
4.2. Methods based on probabilistic models
4.2.1. Naive Bayes
4.2.2. Maximum Entropy.
4.3. Methods based on distances.
4.3.2. Linear classifier based on distances.
4.3.3. Clustering within classes.
4.4. Rules-based methods.
4.4.1. Decision trees.
4.5. Linear classifiers and methods based on Kernels.
4.5.1. Linear classifier based on scalar product
4.5.2. Linear Classifier with Kernel
4.5.3. Kernels for word processing
4.5.4. Support Vector Machines
4.6. Test protocols.
4.6.1. Validation protocols.
4.6.2. Measures of evaluation
4.6.3. Statistical Tests
4.6.4. Comparison of classifiers.
5.1.1. Typology of optimization methods.
5.1.2. Characteristics of optimization metaheuristics
5.2. Optimization using Lagrange multipliers.
5.2.1. Description of the method.
5.2.2. Application example.
5.2.3. Analysis of the method
5.3. Gradient descent
5.3.1. Presentation of the idea
5.3.2. Application example.
5.3.3. Additional questions
5.4. Basin hoping
5.4.1. Description of the method
5.4.2. Example of application
5.4.3. Analysis of the method
5.5. Genetic algorithms.
5.5.1. Description of method
5.5.2. Extensions and improvements.
5.5.3. Examples of application
5.5.4. Compilation of statistics
5.5.5. Combinatory problems
5.5.6. Problems with constrains
5.5.7. Analysis of the method
5.6. Ants colonies
5.6.1. Description of method
5.6.2. Application example
5.6.3. Analysis of the method
5.6.4. Source code in Python
5.7. Optimization with Particle Swarms
5.7.1. Description of method
5.7.2. Application example
5.7.3. Analysis of the method
5.7.4. Source code in Python
5.8. Taboo search.
5.8.1. Description of method
5.8.2. Application example
5.8.3. Analysis of the method
5.8.4. Source code in Python
6. Deep learning
6.1.1. Recent achievements
6.2. Neural networks
6.2.1. Components of a neural network
6.2.2. Activation functions
6.2.3. Training of a neuran network
6.2.4. Learning problems
6.2.5. Some solutions
6.2.6. Deep learning
6.3. Multilayer Perceptron
6.3.2. Example of MLP
6.4. Classification of images with convolutional neural networks (CNN)
6.4.1. Implementation of CNN in Python using the Keras libraries
6.5. Recurrent networks
6.6. Other architectures
6.6.2. Reinforcement Learning
6.6.3. Generative Systems
7. Annex: basic statistical concepts
The subject is composed of didactic modules in paper support, which contain self-assessment exercises with diverse solutions and activities.
This material will be complemented with the one that the consultants put within the reach of the students to the classroom of the subject.
It the classroom there will be a space to support and solve the doubts corresponding to the Python language.
The assessment process is based on students' own work and the assumption that this work is original and has been carried out by them.
In assessment activities, the following irregular behaviours, among others, may have serious academic and disciplinary consequences: someone else being involved in carrying out the student's assessment test or activity, or the work being not entirely original; copying another's work or committing plagiarism; attempting to cheat to obtain better academic results; collaborating in, covering up or encouraging copying; or using unauthorized material, software or devices during assessment.
If students are caught engaging in any of these irregular behaviours, they may receive a fail mark (D/0) for the assessable activities set out in the course plan (including the final tests) or in the final mark for the course. This could be because they have used unauthorized materials, software or devices (e.g. social networking sites or internet search engines) during the tests, because they have copied text fragments from an external source (internet, notes, books, articles, other student's projects or activities, etc.) without correctly citing the source, or because they have engaged in any other irregular conduct.
In accordance with the UOC's academic regulations , irregular conduct during assessment, besides leading to a failing mark for the course, may be grounds for disciplinary proceedings and, where appropriate, the corresponding punishment, as established in the UOC's coexistence regulations.
In its assessment process, the UOC reserves the right to: