Explore our wide array of courses, lectures, and labs. You can find our current offerings listed below. Look forward to exciting and enriching experiences with us!

Lecture: Advanced Methods For Text Mining

This course, offered as part of the Master's Program in Media Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), dives into the sophisticated realm of text mining. Acknowledging text as a prevalent medium of communication, the course grapples with the complexities of textual data mining. Students will embark on a journey through both the foundational theories of data mining and machine learning and the cutting-edge practices in text analysis. From the preliminary discussions on natural language processing to the intricate mechanisms of advanced text mining methods, the course covers a broad spectrum, including Latent Semantic Indexing, Word Embeddings, conventional and resource-efficient Recurrent Neural Networks, Attention Mechanisms, and Transformer architectures. The curriculum extends to practical applications in real-world scenarios, such as natural language inference, information extraction in financial documents, recommender systems in legal and audit sectors, and generative models for digital forensics. A particular focus will be on the exploration of emerging generative language models like GPT-4 and Bloom, highlighting their significance and the nuanced differences that set them apart.

Lecture topics

1. Introduction to Natural Language Processing: Starting from the basics, this lecture introduces the core concepts of NLP and its evolution leading up to the dominance of Transformer models in the machine learning landscape.
    
2. Data Mining and Machine Learning Preliminaries: A foundational overview of data mining techniques and machine learning principles, setting the stage for more advanced studies.
    
3. Advanced Text Mining Methods: Deep dive into sophisticated text mining techniques, including Latent Semantic Indexing, Word Embeddings, and the dynamics of Recurrent Neural Networks for sequential text representation.
    
4. Attention Mechanisms and Transformer Architectures: Exploration of the groundbreaking Attention Mechanism and Transformer models that have revolutionized natural language processing.
    
5. Applications in the Real World: Insight into the application of text mining techniques for natural language inference, information extraction in financial documents, and recommender systems within the legal and audit domains.
    
6. Generative Models for Digital Forensics: Examination of the role of generative models in digital forensics, showcasing their potential for innovation and problem-solving in the digital age.
    
7. Emerging Generative Language Models: A focused discussion on the latest advancements in generative language models, including GPT-4 and Bloom, and their unique contributions to the field of text mining.

Lecture: Mining Media Data I

This course, offered as part of the Master's Program in Media Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), provides a comprehensive exploration of advanced data mining techniques tailored for media data analysis. Students will delve into methods like affinity mining, latent pattern mining, neural networks, and archetypal analysis to uncover insights in behavioral profiling, recommender systems, and outlier detection. Emphasis is placed on theoretical understanding and practical application through mathematical optimization, interpretable models, and real-world case studies, enabling participants to harness data for impactful digital marketing, fraud detection, and content personalization.

Lecture topics

1. Analyze and extract meaningful relationships from large-scale media datasets using advanced data mining techniques.

2. Develop and implement predictive and descriptive models for applications such as recommender systems, trend analysis, and outlier detection.

3. Apply mathematical optimization methods to create interpretable and efficient machine learning models.

4. Integrate theoretical concepts with practical tools to address challenges in digital forensics, behavioral profiling, and marketing strategy design.

Lecture: Mining Media Data II

This course, offered as part of the Master's Program in Media Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), explores advanced techniques in data mining, emphasizing predictive and prescriptive methods applied to media data. Students will learn to analyze large and complex datasets using state-of-the-art machine learning methodologies, including behavioral prediction, knowledge distillation, and large language models (LLMs). The curriculum includes foundational concepts, text representation learning, transformer architectures, and practical applications in media analytics, such as recommendation systems and information extraction. The course combines theoretical instruction with hands-on exercises to develop both technical and analytical skills relevant to industry and research.

Lecture topics

1. Understand and implement advanced data mining techniques for predictive and prescriptive analytics.

2. Employ large language models and transformer-based architectures for tasks like text analysis, classification, and summarization.

3. Apply knowledge distillation techniques to optimize and deploy machine learning models in resource-constrained environments.

4. Analyze media data effectively to derive insights and support decision-making in real-world applications, including digital marketing and fraud detection.

5. Address challenges in media analytics, such as ethical considerations, model interpretability, and efficient resource use.

Lab: Explainable AI and Applications

In this lab “Explainable AI and Applications -- Explainability of foundation models for sequential data”, we will start with the reproduction of existing explainability of deep-learning systems (especially foundation models) in the fields of biomedicine and natural language processing, where foundation models can be thought of an “average mind” for both. Then, we will encourage lab participants to find limitations and explore novel solutions with experiments. The students will work in groups on a selected task.

The lab will be given online via Zoom. We offer the lab course for up to 5 groups with a maximum of four members per group, thus encouraging collaboration and peer learning

Lab Activities

1. Understanding the Landscape: Initiating the course with a comprehensive survey of explainable AI, defining key problems and structuring the foundation for subsequent exploration.
    
2. Reproducing Key Findings: Students will select research papers to reproduce significant findings, applying theoretical concepts practically to affirm groundbreaking work.
    
3. Midterm Milestone: A mid-term presentation provides a platform for students to share their progress, challenges, and insights, enhancing learning through peer feedback.
    
4. Forging New Paths: In the contribution phase, creativity is paramount as groups develop new ideas, benchmarks, datasets, or methodologies to advance the field of XAI.
    
5. Showcasing Innovations: Teams present their original contributions, sharing their innovations with classmates and faculty, fostering an environment of intellectual growth and discovery.
    
6. Reflecting and Reporting: The course concludes with a reflection and reporting phase, documenting the learning journey, findings, and student reflections on the process and outcomes.

This course represents an opportunity for students in the field of machine learning to learn more about the interpretability of AI systems.

Lab: Hybrid Learning an Applications

This lab offers a comprehensive introduction to hybrid learning, merging machine learning and deep learning techniques to address complex problems. By integrating foundation models with downstream tasks using various machine learning methods, students explore a range of fascinating applications. They are encouraged to select and research their own project topics, gaining hands-on experience in data preprocessing, model building, evaluation, and optimization. This course is designed to equip students with practical skills to design and implement effective hybrid learning solutions.

Lab Activities

1. Independent Research and knowledge acquisition: Students should bring their own ideas. They study a self-selected research topic related to hybrid learning, reproduce important the results and elaborate the findings based on their own research. Students need to complete a project within a self-defined scope and timeline.
    
2. Practical Application: Students learn to apply their theoretical knowledge to real-world problems, thus developing a deeper insight into a specific research field. They also become familiar with external research work and how to apply, and adapt relevant aspects to their own projects.
    
3. Communication Skills: The structured presentations in oral and written form helps the students to communicate complex ideas clearly and effectively by following best practices in academic research
    

Seminar: Theory of Deep Learning

This seminar, part of the advanced studies in the Master's Program in Media Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), focuses on the theoretical underpinnings of deep neural networks, particularly exploring the infinite width limit. This limit provides an analytically tractable way to examine neural network properties, enhancing our understanding of aspects like data recognition, generalizability metrics, and structurally significant features of neural networks. Participants will delve into the connection between Gaussian processes, kernel learning, and neural networks; the role of the neural tangent kernel in training; the universal language of tensor programs for architecture-independent insights; and the application of these theories to enhance model generalizability.

Seminar Topics

1. Introduction to Neural Network Theory: Covers the basics of neural network properties under the infinite width limit and its implications for theory and practice.
    
2. Gaussian Processes and Kernel Learning: Examination of the relationship between Gaussian processes, kernel learning, and deep neural networks.
    
3. Neural Tangent Kernel and Training Dynamics: Analysis of how the neural tangent kernel influences neural network training and its implications for learning speed and efficiency.
    
4. Tensor Programs in Neural Networks: Exploration of tensor program language as a model for deriving insights across different neural network architectures.
    
5. Generalizability and Theoretical Applications: Insights into how theoretical advancements translate into improved generalizability in practical applications.

Seminar: Data Science for Medical Applications

Modern medicine is confronted with a multitude of complex data and decisions based on it. Today, high-performance measurement systems and medical case forms provide a wealth of information that needs to be extracted. Examples: genomic sequence data, X-ray images, MRI. In the seminar, current, relevant research progress in the field of data science in the analysis of medical data is reviewed and presented by the students.

Seminar Topics

1. The students familiarize themselves with different types of biomedical data (e. g. medical image data, genomic data, survey data). In the seminar they explore the benefits data science can offer to medicine.
    
2. They read about case studies where data science has been successfully integrated into clinical workflows to improve patient care and they identify barriers for the adoption of data science in healthcare settings and propose strategies to overcome these challenges.
    
3. They get an overview about the latest research and technological advancements in the medical field.
    
4. Students enhance their critical thinking and problem-solving skills by addressing complex medical data challenges (e. g. protein folding, gene regulation) and they understand the ethical and privacy considerations in handling sensitive medical data.