S1: Exploration of ideas and delimitation of the dissertation problem: analysis of research topics in AML, literature review and identification of gaps.
S2: Definition of the theoretical and methodological framework: selection of tools, experimental design, data collection and preparation, definition of evaluation metrics.
S3: Development of the research project: implementation of ML models or algorithms, empirical tests, comparison of approaches, interpretation of results and practical relevance.
S4: Organizing and writing the dissertation: structure, standardizing references, writing sections (introduction, literature review, methodology, results, conclusions).
S5: Presentation and discussion of partial results: tutorial sessions and online seminars, with feedback from the supervisor and peers.
S6: Preparation for public defense: fine-tuning of results, final writing, presentation and argumentation training in videoconference or in person.

The assessment of the curricular unit “Dissertation in Applied Machine Learning” is based on the completion of an original piece of research, a written report (dissertation) and its public defense before a jury, in accordance with the regulations in force at ISCTE.
The dissertation must demonstrate scientific and methodological rigor, relevance and applicability in the field of AML, as well as a clear and coherent structure, including an introduction, literature review, methodology, tests, results and conclusions.

Assessment elements:
- Written document (Dissertation): The scientific and technical quality of the work, the ability to formulate a research problem, the relevance of the hypotheses/research questions, the collection and analysis of empirical data, the application of Machine Learning models, the discussion of the results and the conclusions reached are assessed. Also valued are the clarity and consistency of the text, respect for ethical and deontological standards, practical and/or theoretical relevance and the adequacy of bibliographical references.
- Public defense: The student presents and discusses the results of their dissertation in front of a jury, demonstrating knowledge, the ability to synthesize and respond critically to technical or conceptual issues. The way in which the student organizes the presentation, the quality of the supports used (slides, demonstrations) and the soundness of the arguments during the debate are also weighted.

The grade obtained in the dissertation and defense follows the scale in force at ISCTE, and can range from a qualitative mention of “Fail” to “Pass with distinction” (or equivalent), and obeys the rules of 20 points for the Master's degree.

There will be no additional written exam, as this is an investigative curricular unit.

If the dissertation is not approved, the student will be considered to have failed and may reformulate the work, in accordance with the applicable regulations.

The presentation and defense can take place in the 1st or 2nd Season, in accordance with the academic calendar.

Students with special status should consult the specific provisions regarding the extension of deadlines and defense conditions in Iscte's Regulations for Students with Special Status (REEE).

In addition to the RGACC, students should also consult the Code of Academic Conduct (CCA).

Mandatory

1. Grigoroudis, E. (2023). Handbook of research on artificial intelligence, innovation and entrepreneurship (E. G. Carayannis & E. Grigoroudis, Eds.). Edward Elgar Publishing. 2. Naseem, S. B. (2021). The book towards research in AI & ML: A curated compilation of research chapters from the recent trends in artificial intelligence and machine learning. 3. Hassanien, A. E., Darwish, A., & Chowdhary, C. L. (Eds.). (2019). Handbook of research on deep learning innovations and trends. IGI Global.

Optional

Given the diversity of professional contexts and business challenges where projects can be developed, the program presents guiding topics that are refined as the student builds and implements their project:
S1: Observation and analysis of problems and opportunities for applying ML in business or institutional environments.
S2: Planning and developing research and/or development-intervention projects, defining clear objectives and work stages.
S3: Methodologies and strategies for analyzing requirements, designing and implementing AML solutions adjusted to real contexts.
S4: Operationalizing research and development methodologies, carrying out empirical tests, analyzing results and evaluating impact in the chosen professional context.
S5: Writing reports and scientific/technical texts, writing professional documentation, good practices in managing code, data and results.

The assessment of the curricular unit “Project Work in Applied Machine Learning” is based on the completion of an original piece of research, a written report (Project Work Report - PWP) and its public defense before a jury, in accordance with the regulations in force at ISCTE.
The PWP must demonstrate scientific and methodological rigor, relevance and applicability in the field of AML, as well as a clear and coherent structure, including an introduction, literature review, methodology, tests, results and conclusions.

Assessment elements:
- Written document (PWP): The scientific and technical quality of the work, the ability to formulate a research problem, the relevance of the hypotheses/research questions, the collection and analysis of empirical data, the application of Machine Learning models, the discussion of the results and the conclusions reached are assessed. Also valued are the clarity and consistency of the text, respect for ethical and deontological standards, practical and/or theoretical relevance and the adequacy of bibliographical references.
- Public defense: The student presents and discusses the results of their PWP in front of a jury, demonstrating knowledge, the ability to synthesize and respond critically to technical or conceptual issues. The way in which the student organizes the presentation, the quality of the supports used (slides, demonstrations) and the soundness of the arguments during the debate are also weighted.

All assessment elements must be carried out individually.

The grade obtained in the PWP and defense follows the scale in force at ISCTE, and can range from a qualitative mention of “Fail” to “Pass with distinction” (or equivalent), and obeys the rules of 20 points for the Master's degree.

There will be no additional written exam, as this is an investigative curricular unit.

If the PWP is not approved, the student will be considered to have failed and may reformulate the work, in accordance with the applicable regulations.

The presentation and defense can take place in the 1st or 2nd Season, in accordance with the academic calendar.

Students with special status should consult the specific provisions regarding the extension of deadlines and defense conditions in Iscte's Regulations for Students with Special Status (REEE).

In addition to the RGACC, students should also consult the Code of Academic Conduct (CCA).

Mandatory

1. Grigoroudis, E. (2023). Handbook of research on artificial intelligence, innovation and entrepreneurship (E. G. Carayannis & E. Grigoroudis, Eds.). Edward Elgar Publishing. 2. Naseem, S. B. (2021). The book towards research in AI & ML: A curated compilation of research chapters from the recent trends in artificial intelligence and machine learning. 3. Hassanien, A. E., Darwish, A., & Chowdhary, C. L. (Eds.). (2019). Handbook of research on deep learning innovations and trends. IGI Global.

Optional

S1: Introduction to scientific research and research methodology in AI and ML.
S2: Identification and choice of research themes and topics in machine learning and deep learning, considering interdisciplinary scenarios.
S3: Design of research projects: definition of objectives, state of the art, theoretical foundation, formulation of hypotheses, methodological selection (qualitative, quantitative or mixed), data collection and analysis, structure and format of the work plan.
S4: Practical exercises in preparing research proposals: oral presentation of chosen topics, critical discussion in class or in forums, peer review and feedback.
S5: Ethics and deontology in ML research: issues of privacy, transparency, explainability and algorithmic bias.
CP6: Writing scientific texts: writing standards, bibliographic referencing, standardization of citations, preparation of research reports for dissertations/master's projects.

The assessment follows the Assessment Throughout the Semester (ALS) model set out in the RGACC and is based on continuous monitoring of student progress throughout the semester, reflecting the coherence between the learning process and the results obtained.
There is no final exam, and the emphasis is on practical mini-deliverables and oral discussions:

The assessment elements are as follows
- 2 Quizzes and/or 2 formative activities (20%)
Quick tasks (e.g. online quizzes, micro-reports) are offered to gauge understanding of ML research concepts and individual progress.
These quizzes serve as a diagnostic and provide immediate feedback on possible gaps and difficulties.

- Mini-research project (50%): The core element of the assessment is the drafting of a research plan (4-6 pages), which includes:
-- Definition of the research theme/topic and respective objectives;
-- A summary of the state of the art, justifying the relevance and framework of the work;
-- Proposed methodology(ies), including data collection and analysis, validation of results and resource management;
-- Ethical and deontological considerations;
-- Timing and structure of the report/dissertation.
The quality of the document, methodological rigor, bibliographical relevance, logical coherence and clarity of arguments are assessed.
A minimum mark of 10 is required to pass this component.

- Oral discussion of the mini-project (30%)
Each student presents their mini-research project orally (via videoconference), followed by a question and answer period.

Communication skills, clarity and structure of speech, ability to answer technical or conceptual questions and articulation with suggestions from colleagues and the lecturer are assessed.

The assessment elements “Mini-research project” and “Mini-research project oral discussion” are compulsory to pass the curricular unit.

A student is considered to have failed if he or she does not pass the semester-long assessment (quizzes + mini-project + discussion) for this curricular unit.

The grade can be improved in the following academic year by repeating the assessment components, in accordance with the regulations in force.

The 1st Season and 2nd Season can be used for assessment.

Attendance at synchronous sessions is not compulsory.

In addition to the RGACC, we recommend reading other reference documents for the assessment process, such as the Regulations for Students with Special Status (REEE) and the Code of Academic Conduct (CCA).

Mandatory

1. Grigoroudis, E. (2023). Handbook of research on artificial intelligence, innovation and entrepreneurship (E. G. Carayannis & E. Grigoroudis, Eds.). Edward Elgar Publishing. 2. Naseem, S. B. (2021). The book towards research in AI & ML: A curated compilation of research chapters from the recent trends in artificial intelligence and machine learning. 3. Hassanien, A. E., Darwish, A., & Chowdhary, C. L. (Eds.). (2019). Handbook of research on deep learning innovations and trends. IGI Global.

Optional

S1: Introduction to deep learning: key concepts, history, distinction from traditional machine learning.
S2: Convolutional, dense, recurrent layers, normalization, dropout, activation functions and regularization techniques.
S3: Selection and application of appropriate techniques for different types of data such as 2D/3D images, univariate and multivariate time sequences.
S4: Practical introduction to the TensorFlow, Keras and PyTorch frameworks, development of simple models and analysis of results.
S5: Pre-existing architectures (e.g., ResNet, VGG, U-Net) and their reuse, transfer of learning and fine-tuning of pre-trained models.
S6: Performance metrics, interpretation of results, identification of bottlenecks, optimization proposals and fine-tuning.
S7: Exploration of real case studies, from the definition of the challenge, data modeling, implementation and training of the model, to critical analysis and presentation of the results.

Due to its practical nature, this curricular unit follows the model of ssessment throughout the semester (ALS, provided for in the RGACC) by project, without a final exam, and includes the following assessment components:

-- Mini-Project (30%):
In this mini-project, carried out in groups, students develop a simple deep learning application, from data preparation to the first functional version of the model. The assessment focuses on implementation skills, code clarity, and the correct use of the tools covered in the course. It also includes an individual oral presentation as part of the grading.

-- Final Project (70%):
The Final Project, carried out in groups, is more complex and requires the integration of layers, architectures, and advanced techniques. Emphasis is placed on model optimization, critical analysis of results, and the ability to propose improvements. This project allows students to consolidate all the skills they have acquired. It also includes an individual oral presentation as part of the grading.

Each assessment component has a minimum mark of 8.5.

For any assessment component, the lecturer or course coordinator may, at any time during the semester, request an individual oral discussion to confirm the authorship of the work submitted and/or to assess the knowledge and skills demonstrated by the student.

The use of generative artificial intelligence tools is permitted solely as a support tool for the learning process; under no circumstances may they replace the student’s intellectual authorship and individual responsibility for any assessment work submitted. The student bears full responsibility for the content submitted and must be able to justify, explain and defend it before the lecturer and/or coordinator. For further clarification, it is essential to read other reference documents relating to the assessment process, such as the Guide to the Responsible Use of Artificial Intelligence in Academic Work (Scientific Council, 21 April 2026), the Code of Academic Conduct (CCA) and the Student Disciplinary Regulations (RDD).

To obtain approval for this curricular unit, students must achieve a minimum final mark of 10, calculated as the weighted sum of all assessment components.

A student is considered to have failed this curricular unit if they do not obtain approval in the ALS.

The 1st Term and 2nd Term can be used for assessment.

Attendance at synchronous sessions is not compulsory.

Mandatory

1. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press.
2. Godoy, D. V. (2022). Deep learning with PyTorch step-by-step: A beginner’s guide: Volume I: Fundamentals.
3. Chollet, F. (2021). Deep Learning with Python (2.ª ed.). Manning.

Optional

4. Stevens, J., Antiga, L., & Viebranz, T. (2020). Deep Learning with PyTorch. Manning.
5. Patterson, J., & Gibson, A. (2017). Deep Learning: A Practitioner’s Approach. O’Reilly."

PC1: Image representation.
PC2: Image operations.
PC3: Image feature extraction.
PC4: Introduction to machine learning (revision).
PC5: Artificial neural networks (revision).
PC6: Architecture and operation of convolutional neural networks (CNNs).
PC7: Transfer learning.
PC8: CNN architectures for object detection and identification.
PC9: Network architectures for automatic multimedia content generation.
PC10: Development of practical applications in computer vision and machine learning.

Assessment will be carried out throughout the semester and will consist of the following elements:
- Participation in synchronous sessions (10%, individual);
- Completion of asynchronous training activities (10%, individual);
- Work assignments and final oral discussion (80%, carried out in groups but with individual assessment)*.

*The "Work assignments" component is subject to an oral discussion. If the knowledge demonstrated by the student in the discussion is noticeably below the quality of the submitted work, the discussion will impose a limit on the grade for this component (the student may even fail the course if their performance in the discussion is not minimally satisfactory).

Mandatory

"Feature Extraction and Image Processing for Computer Vision, 4th Edition, M. Nixon e Alberto Aguado, Academic Press, 2019 Deep Learning, I. Goodsfellow, Y. Bengio e A. Courville, MIT Press, 2016 Learning OpenCV 4 with Python 3, 3rd Edition, Joseph Howse, Joe Minichino, Packt Publishing, 2020 Tutoriais e documentação das bibliotecas OpenCV e Tensorflow

Optional

S1. Fundamental concepts of LLMs and the Transformer architecture, including attention mechanisms and positional encoding.
S2. Pratical application of LLMs using tools such as Hugging Face and PyTorch for creating and optimizing NLP models.
S3. Methods for training and fine-tuning models for personalization in different contexts.
S4. Implementation of solutions for tasks such as named entity recognition, sentiment analysis, text generation, summarization, and translation.
S5. Appropriate metrics to evaluate the performance of models in NLP tasks.
S6. Retrieval-Augmented Generation (RAG) models to enhance LLM performance.
S7. Visual language models and exploration of tasks such as image classification, object detection, and description.

This course uses the assessment throughout the semester modality, not contemplating the mode of assessment by exam. The evaluation consists of 2 components:

1. Group assignments carried out throughout the semester:
- Presentation of a scientific paper (5%).
- Fine-tuning an LLM for a specific field (10%).
- Development of a case study with correct evaluation of the model (40%).
- Creation of a model to improve the LLM developed in assignment 2 (25%).
2. Test (20%).

The average of the group work and the test requires a minimum mark of 9.5.
Grades can only be improved by reattempting the assessment process in the subsequent year.
Attendance at synchronous sessions is not mandatory.

Mandatory

"Dan Jurafsky and James H. Martin. Speech and Language Processing (2024 pre-release) (https://web.stanford.edu/~jurafsky/slpdraft/) Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning (https://www.deeplearningbook.org/) Delip Rao and Brian McMahan. Natural Language Processing with PyTorch Lewis Tunstall, Leandro von Werra, and Thomas Wolf. Natural Language Processing with Transformers (https://transformersbook.com/)

Optional

CP1: Fundamentals of parallel computing and GPU architectures. Parallelism, concurrency, scalability, threads, blocks, grids and GPU memory hierarchy.
CP2: Parallel GPU programming using CuPy, Numba and CUDA. Kernel development, memory management, compilation and optimization of parallel operations.
CP3: GPU performance analysis and optimization. Profiling, computational bottleneck identification, memory optimization and efficient kernel configuration.
CP4: Frameworks for accelerated and distributed computing. RAPIDS, Dask and integration with PyTorch for deep learning.
CP5: Advanced acceleration techniques for deep learning. Mixed Precision, Gradient Accumulation, Activation Checkpointing and GPU optimization.
CP6: Case studies and development of high-performance computing applications for deep learning and computer vision.

As this curricular unit is very practical and application-oriented, it follows the semester-long assessment model (ALS, provided for in the RGACC) by project, without a final exam, and includes the following assessment components:

- Project (100%):
The project consists of developing a practical assignment that integrates the knowledge and skills acquired throughout the semester.

The Project assessment (100%) will be conducted in phases as follows:
-- 1st Tutorial (20%): In this phase, the student presents the initial project plan, including the problem definition and the objectives to be achieved through the project.
-- 2nd Tutorial (30%): In this second phase, the focus is on the technical progress of the project.
-- Final Tutorial (50%): In the final phase, the student presents the final version of the project, including all functional code, a detailed performance analysis, a discussion of the results obtained, the difficulties encountered, and the solutions adopted. It also includes an individual oral presentation as part of the assessment.

The three tutorials combined (1st, 2nd, and Final) account for 100% of the grade, and each tutorial has a minimum passing grade of 8.5.

For any assessment component, the lecturer or curricular unit coordinator may, at any time during the semester, request an individual oral discussion to confirm the authorship of the work submitted and/or to assess the knowledge and skills demonstrated by the student.

The use of generative artificial intelligence tools is permitted solely as a support tool for the learning process; under no circumstances may they replace the student’s intellectual authorship and individual responsibility for any assessment work submitted. The student bears full responsibility for the content submitted and must be able to justify, explain and defend it before the lecturer and/or coordinator. For further clarification, it is essential to read other reference documents relating to the assessment process, such as the Guide to the Responsible Use of Artificial Intelligence in Academic Work (Scientific Council, 21 April 2026), the Code of Academic Conduct (CCA) and the Student Disciplinary Regulations (RDD).

To obtain approval for this curricular unit, students must achieve a minimum final mark of 10, resulting from the weighted sum of all tutorial sessions.

A student is considered to have failed this curricular unit if they do not obtain approval in the ALS.

The 1st Term and 2nd Term can be used for assessment.

Attendance at synchronous sessions is not compulsory.

Mandatory

1. Tuomanen, B. (2018). Hands-On GPU Programming with Python and CUDA: Explore high-performance parallel computing with CUDA. Packt Publishing.
2. 1. Kirk, D. B., & Hwu, W.-M. W. (2016). Programming Massively Parallel Processors: A Hands-on Approach
3. Storti, D., & Yurtoglu, M. (2016). CUDA for engineers: An introduction to high-performance parallel computing (1st ed.). Addison-Wesley Professional.
4. Wilt, N. (2013). The CUDA Handbook: A comprehensive guide to GPU programming (1st ed.). Pearson Education.

Optional

5. Cook, S. (2013). CUDA Programming: A Developer’s Guide to Parallel Computing with GPUs. Morgan Kaufmann.
6. Sanders, J., & Kandrot, E. (2011). CUDA by Example: An Introduction to General-Purpose GPU Programming. Addison-Wesley.

S1: Fundamentals of Intelligent Robotics, definition, basic concepts and applications; Main components of an intelligent robotic system and types of autonomous robots; Emerging technologies in robotics and artificial intelligence.
S2: Introduction to ROS; Navigation in simulated environments; Practical implementation of navigation algorithms using simulations.
S3: Sensory Perception and Interpretation in Robotics; Types of sensors used; Methods for acquiring and processing sensory data; Introduction to SLAM.
S4: Supervised, evolutionary, reinforcement and deep learning for robotics. Applications for object recognition, navigation and autonomous decision-making.
S5: Human-Robot Interaction and Collaboration: Fundamentals, communication interfaces, feedback and control.
Intuitive communication methods; collaborative robots (cobots).
S6: Design and Testing of Integrated Robotic Systems

The assessment of this curricular unit follows the model of assessment throughout the semester, favoring regular monitoring of student progress, as well as the development of a final project that consolidates the knowledge and skills acquired.
It has the following composition

- 4 Weekly Challenges (40%):
Throughout the semester, weekly challenges will be proposed based on mini-projects, to be submitted on the e-learning platform, which aim to assess understanding and application of the topics covered.
Each challenge will be corrected with formative feedback, allowing the student to continually improve their understanding of the content.
The final average of all the weekly challenges must be equal to or greater than 9.5 points. Otherwise, the student will be considered to have failed this component.
The weekly challenges can be done individually or in groups.

- Final Project (60%):
The final project, carried out in groups, consists of developing an integrated robotic system.
Students must design and implement a coherent solution, prepare a technical report describing the approach, the methodologies used, the results obtained and the difficulties encountered, and make an oral presentation (online) of the work, justifying the options adopted and discussing potential improvements.
The final project has a minimum mark of 9.5.

In accordance with the RGACC, the student can opt for assessment by exam (in the 1st season, 2nd season or special season).
In this case, the final project is worth 100% of the grade, replacing the weekly challenges.
The project must follow the same guidelines, including a technical report and oral presentation, and the minimum mark of 9.5 for approval is also applied.

To pass the curricular unit, the student must obtain a final mark of at least 10 points, taking into account the weighting of the weekly challenges (40%) and the final project (60%).
If they do not pass, they will be considered to have failed.

Attendance at synchronous sessions is not compulsory.

In addition to the RGACC, students should consult the Regulations for Students with Special Status (REEE) and the Code of Academic Conduct (CCA).

Mandatory

"1. Thrun, S., Burgard, W.,, Fox, D. (2005). Probabilistic robotics. Cambridge, Mass.: MIT Press. ISBN: 0262201623 9780262201629 2. Chernova, S., & Thomaz, A. L. (2014). Robot learning from human teachers. Springer. 3. Craig, 1955. (1986). Introduction to robotics : mechanics & control / John J. Craig.. Reading, Mass.: Addison-Wesley Pub. Co.,. ISBN: 0201103265

Optional

S1: Concepts and fundamentals of GANs, basic architectures (DCGAN, WGAN), training process (generator vs. discriminator), challenges and practical applications (image generation, synthetic data).
S2: Review of reinforcement learning concepts, fundamental algorithms (DQN, PPO, AlphaZero), simulated environments, exploration vs. exploitation.
S3: Time series analysis with neural networks: LSTMs, GRUs and transformers for sequential data, particularities of temporal modeling (seasonality, trend), evaluation of forecasts and improvement strategies.
S4: Principles of deep evolutionary algorithms, genetic operators, adaptation strategies, use cases in hyperparameter optimization.
S5: Interpretation of advanced models, explainability and fairness techniques to promote transparency and robustness in deep models.
S6: Development of practical projects in areas such as health, banking, finance or IoT, integrating multiple advanced techniques.

The evaluation follows the model of Evaluation Throughout the Semester (foreseen in the RGACC) 100% by project/practical work, without a final exam, promoting the continuous involvement of the student. The assessment components are:

- Three mini-works (20% each)
Throughout the semester, each mini-work focuses on a specific topic:
-- Creation of a simple GAN, focusing on defining the architecture and training the generator and -discriminator.
-- Deep Reinforcement Learning model, including the choice of algorithm (DQN, PPO, etc.) and the validation of training in a simulated environment.
-- Time Series Modeling, using LSTMs, GRUs or transformers, evaluating their forecasting capacity on sequential data.
The practical implementation, the ability to analyze the results, the clarity of the code and the correct use of the libraries are evaluated in each mini-work.
The mini-works can be done individually or in groups.
There is no minimum mark for each mini-work, but the whole represents 60% of the final mark.

- Final Project (40%)
In a larger assignment, students can integrate one or more topics from the mini-assignments (GANs, Reinforcement Learning, Time Series, Deep Genetic/Evolutionary Algorithm), or explore a different scenario that requires a combination of advanced techniques. Optimization of the model, critical analysis of results, formulation of improvements and the ability to justify the technical choices made are valued. This project requires the submission of a technical report and a presentation (online) to defend the methods and conclusions, allowing communication skills and mastery of the subject to be assessed.
The final project must be carried out in groups.

The final grade results from the weighted sum of the three mini-works (20% each) and the final project (40%).

The final grade must be at least 10 points.

If the student does not achieve this mark, they are considered to have failed.

The grade can be improved in the following academic year by repeating the assessment components in accordance with the regulations in force.

Assessments can take place in the 1st and 2nd Seasons, and there is no additional final exam.

Attendance at synchronous sessions is not compulsory, but strongly recommended for the richness of the discussions and the immediate feedback on individual and collective progress.

It is also recommended to read the Regulations for Students with Special Status (REEE) and the Code of Academic Conduct (CCA).

Mandatory

Optional

S1: Classical numerical methods, derivatives and gradients, unconstrained and constrained optimization, introduction to convergence analysis.
S2: Simulated Annealing Metaheuristics, Tabu Search, Evolutionary Algorithms; applications to complex problems (model selection, hyperparameter tuning, global optimization search).
S3: Gradient descent (SGD, Momentum, Adagrad, Adam); implementation in deep neural networks; analysis of convergence and stability graphs of training and validation processes.
S4: Loss/cost functions in Deep Learning: impact on convergence, generalization capability, and model robustness.
S5: Comparison of methods, stopping criteria, scalability, result analysis, improvement proposals, and discussion of practical cases.

As this curricular unit is of a very practical and application-oriented nature, it follows the semester-long assessment model (provided for in the RGACC), and therefore does not have a final exam.
The assessment of this UC has the following composition:

- 2 Quizzes (10% each, total 20%):
Individual short online quizzes, distributed throughout the semester, to assess the theoretical understanding of optimization concepts, metaheuristics, gradient descent methods, and cost functions. There is no minimum score on the quizzes, but completing them helps consolidate knowledge and provides immediate feedback to the student.

- Final Project (80%):
The final project, carried out in groups, is developed in two phases (tutorials), reflecting the PBL (Project-Based Learning) approach:

-- 1st Tutorial (20%):
Delivery of a work plan, identifying the optimization problem in AI, the techniques to be used, as well as the evaluation criteria and convergence metrics.

-- 2nd Tutorial (60%):
Online submission and defense of the complete project, including code, experimental results, critical performance analysis, method comparison, and improvement proposals.

A minimum score of 9.5 is required for the final project as a whole.

The final grade results from the weighted sum of the quizzes (20%) and the final project (80%).

For approval, it is necessary to obtain at least 10 values.

If the student does not pass, they may retake the UC in the following academic year, according to the regulations.

The improvement of the grade can be obtained in the following academic year by repeating the assessment components, in accordance with the current regulations.

The 1st Period and 2nd Period may be used for conducting assessment moments.

Presence during synchronous moments is not mandatory, but recommended, given the opportunity for interaction and immediate feedback.

In addition to the RGACC, the student must consult the Regulation for Students with Special Status (REEE) and the Academic Code of Conduct (CCA).

Mandatory

"1. Sra, Suvrit, Sebastian Nowozin, and Stephen J. Wright, eds., Optimization for machine learning, Mit Press, 2012. 2. Dréo, Johann, et al., Metaheuristics for hard optimization: methods and case studies, Springer Science \& Business Media, 2006. 3. Postek, Krzysztof and Zocca, Alessandro and Gromicho, Joaquim and Kantor, Jeffrey, Hands-On Mathematical Optimization with AMPL in Python, 2024. https://ampl.com/mo-book 4. Dive into DeepLearning, Chapter 11, Optimization algorithms, https://d2l.ai/chapter_optimization/"

Optional

"J. A. E. Andersson, J. Gillis, G. Horn, J. B. Rawlings and Moritz Diehl, CasADi -- A software framework for nonlinear optimization and optimal control, Mathematical Programming Computation, vol. 11, no. 1, pp. 1--36, 2019. https://doi.org/10.1007/s12532-018-0139-4, https://web.casadi.org/"

S1: MLOps and LLMOps foundations. Intelligent model lifecycle, differences from DevOps and methodological framing through CRISP-ML(Q).
S2: Experiment management, configuration and reproducibility. MLflow Tracking, DVC and Hydra for versioning data, artefacts, parameters and results.
S3: Orchestration and automation of machine learning workflows. Development of Apache Airflow DAGs for ingestion, training, evaluation and reactive pipeline execution.
S4: Model lifecycle management. Packaging, Model Registry, versioning, traceability, governance and observability of artefacts and experiments.
S5: Model serving, deployment and monitoring. API development with FastAPI, containerisation with Docker, performance monitoring and degradation detection with Evidently.
S6: Introduction to LLMOps. Prompt management and evaluation, tracing, observability and integration of language models into operational pipelines.

Due to its eminently practical and applied nature, this curricular unit follows the project-based Assessment Throughout the Semester (ALS, as provided for in the RGACC) model, does not include exam-based assessment, and consists of the following assessment components:

-- Mini-Project 1 (40%):
Consists of implementing a reproducible pipeline for developing machine learning models, including experiment management, data and model versioning, project configuration, and partial automation of the training cycle.

-- Mini-Project 2 (60%):
This involves evolving the previous solution into a complete system for operationalizing intelligent models, integrating deployment, serving, monitoring, observability, and LLMOps principles (if applicable).

Each assessment component has a minimum mark of 8.5.

For any assessment component, the lecturer or curricular unit coordinator may, at any time during the semester, request an individual oral discussion to confirm the authorship of the work submitted and/or to assess the knowledge and skills demonstrated by the student.

The use of generative artificial intelligence tools is permitted solely as a support tool for the learning process; under no circumstances may they replace the student’s intellectual authorship and individual responsibility for any assessment work submitted. The student bears full responsibility for the content submitted and must be able to justify, explain and defend it before the lecturer and/or coordinator. For further clarification, it is essential to read other reference documents relating to the assessment process, such as the Guide to the Responsible Use of Artificial Intelligence in Academic Work (Scientific Council, 21 April 2026), the Code of Academic Conduct (CCA) and the Student Disciplinary Regulations (RDD).

To obtain approval for this curricular unit, students must achieve a minimum final mark of 10, calculated as the weighted sum of all assessment components.

A student is considered to have failed this curricular unit if they do not obtain approval in the ALS.

The 1st Term and 2nd Term can be used for assessment.

Attendance at synchronous sessions is not compulsory.

Mandatory

"1. Washizaki, H., Prikladnicki, R., et al. (2022). MLOps Engineering at Scale. O’Reilly. 2. Zweig, K. (2023). Effective MLOps. Manning. 3. Villela, J. A., et al. (2022). Kubeflow Operations Guide. O’Reilly. 4. Lakshmanan, V., Robinson, C., & Munn, M. (2020). Machine Learning Design Patterns: Solutions to Common Challenges in Data Preparation, Model Building, and MLOps. O’Reilly. "

Optional

"1. Amershi, S., et al. (2019). Software Engineering for Machine Learning: A Case Study. IEEE/ACM Joint Conference on SE 2. Shevat, A. (2020). Designing ML-Powered Applications: ML Design Patterns for Developer Teams. O’Reilly."

S1: Brief history of AI: historical framework, evolution of methods, trends and socio-economic contexts.
S2: How AI works: technical fundamentals, machine learning approaches, and intelligent systems.
S3: Bias, Ethics, Equity, Privacy, Robustness and Trust: critical analysis of the social, professional and legal implications.
S4: Legal and regulatory frameworks: standards, national and international laws, regulatory bodies and compliance standards.
S5: AI in the present: case studies in diverse fields (health, transport, education, security, media).
S6: Approaches and techniques to make models and decisions more transparent - XAI (Explainable AI)
S7: AI of the future: emerging topics, foresight, social and technical impact scenarios, public participation and accountability.

The evaluation, following the provisions of the RGACC, uses the typology of evaluation throughout the semester, and has the following composition:

(a) [30%] Group work in workshops:
- Development of a practical case study, presented and discussed in an online environment (forums or synchronous sessions).
(b) [20%] Individually evaluate the result of the work submitted in (d)
(c) [30%] Activities with online individual assessments (quizzes)
(d) [20%] Case study summary document (in groups)

It requires a minimum grade of 8 in all four components.

For approval, the student must obtain a final grade ≥ 10 points. Otherwise, the student fails.

Failure to pass the semester allows the option for evaluation by final exam. The exam will focus on all the contents covered (theoretical and practical), consisting of a written test and, if necessary, a complementary oral test to clarify doubts. The final exam has a weighting of 100%.

Admission to the exam is possible for those who did not pass the type of assessment throughout the semester or opted for this assessment regime, under the terms of the RGACC.

The grade improvement can be obtained by repeating the final exam in the following season, in accordance with the regulations.

Attendance at synchronous sessions is not mandatory, but recommended, due to the opportunity for direct interaction and immediate feedback.

In addition to the RGACC, the student must consult the Regulation of Students with Special Status (REEE) and the Code of Academic Conduct (CCA).

Mandatory

"1. T. Miller. Explanation in artificial intelligence: Insights from the social sciences. Artificial Intelligence, 267, 1–38, 2019. 2. C. Molnar. Interpretable machine learning. A Guide for Making Black Box Models Explainable, 2019. https://christophm.github.io/interpretable-ml-book/. 3. A. Holzinger, P. Kieseberg, E. Weippl & A Min Tjoa. Current Advances, Trends and Challenges of Machine Learning and Knowledge Extraction: From Machine Learning to Explainable AI. Springer Lecture Notes in Computer Science LNCS 11015. Cham: Springer, pp. 1-8, 2018. 4. S. Russell, P. Norvig. Artificial Intelligence: A Modern Approach. Pearson, Upper Saddle River, NJ, 2009."

Optional

"1. ALLEA The European Code of Conduct for Research Integrity, European Union, URL: https://allea.org/code-of-conduct/ 2. Artificial intelligence Act, URL: https://www.europarl.europa.eu/RegData/etudes/BRIE/2021/698792/EPRS_BRI(2021)698792_EN.pdf"