1. Introduction to non-structured data. This session provides an overview of the course.
2. Co-occurrence analysis, visualizing high-dimensional data with PCA. In this session frequency and patterns of paired elements (e.g. keywords and codes within a dataset) are analyzed to reveal associations and structural relationships among data components. This session also covers projections of multi-feature datasets into lower-dimensional spaces, enabling clearer interpretation of data structure and variance into two or three dimensions.
3. PCA (cont'd), manifold learning. In this session a family of non-linear algorithms is analyzed in order to uncover low-dimensional structures embedded within high-dimensional data, preserving intrinsic geometric relationships to reveal complex patterns
4. Clustering, k-means, Gaussian mixture models. This session focuses on unsupervised clustering algorithm that partitions data into a predefined number of clusters, and later it is extended by modelling data as a combination of multiple Gaussian distributions.
5. Clustering (cont’') - interpreting GMMs, automatically selecting the number of clusters. In this session automatic cluster count selection is achieved through likelihood-based criteria (e.g. Bayesian Information Criteria), which balances model fit against complexity.
6. Unstructured Data review. The session revisits the challenges of handling unstructured data and specifically those regarding the language or the images, its prevalence in real-world scenarios, and the importance of analytical techniques and the expected meaningful insights. Neuroscience theory is also discussed and foundational concepts, common preprocessing steps, and tools for managing unstructured data.
7. Rule based NLP. This session is focused in rule-based NLP methods, which use handcrafted rules and patterns to analyze and manipulate text. Most essential techniques are coved including tokenization, part-of-speech tagging, named entity recognition, and syntax parsing. Through examples, participants learn the strengths and limitations of rule-based approaches and their role in niche applications or as complements to data-driven models.
8. Neural Networks. This session is intended to refresh the students’ skills about neural networks and prepare them for advanced NLP and deep learning topics, this session revisits the fundamentals of neural networks. Key concepts like perceptrons, activation functions, backpropagation, and architectures such as feedforward, convolutional, and recurrent networks are reviewed.
9. NLP with Deep Learning. In this session, the deep-learning based approaches to NLP will be introduced: how neural networks can handle complex language tasks like sentiment analysis, machine translation, and question answering. Techniques like recurrent neural networks (RNNs), long short-term memory (LSTM), and gated recurrent units (GRUs) are explained, along with their advantages over rule-based methods. Some of these techniques will be demonstrated.
10. Embeddings and vectorization. This session will cover the representation of text data into numbers for machine learning. Word embeddings like Word2Vec, GloVe, and contextual embeddings from models such as BERT will be introduced. Students will learn how vectorization techniques capture semantic relationships and contextual information, enabling sophisticated language modeling and understanding.
11. Transformers and Generative AI. The evolution of transformers will be explored in this session. Concepts like self-attention and multi-head attention mechanisms are explained and also the standard architecture of transformers and how does a transformer work. Models like BERT and GPT will be also introduced as well as different examples.
12. Generative AI and Business applications. This session focuses on the practical applications of generative AI in business. Case studies are presented, demonstrating how AI can enhance customer experiences, optimize workflows, and enable innovative solutions. Ethical considerations, challenges, and best practices for deploying generative AI systems in real-world scenarios are also discussed. RAGs and Agentic RAGs will be covered as well in this session
13. Image processing. From CNN to transformers. This is the closing session of the module and covers both the basics of the image processing, starting with convolutional neural networks (CNNs) and their role in tasks like image classification and object detection. The transition to transformer-based architectures in computer vision is covered, showing how these models have outperformed traditional CNNs in tasks requiring contextual understanding. Second part of the session will be reserved for the final Exam.