VII Course on Machine Learning and Central Banking (theoretical)

The theoretical component of the VII Course on Machine Learning and Central Banking was held virtually on August 24–25, 2026, as part of a training initiative jointly organized by the Center for Latin American Monetary Studies (CEMLA) and the Deutsche Bundesbank. The course aimed to strengthen participants’ knowledge of machine learning techniques and their applications in finance and central banking, providing a conceptual foundation for understanding the opportunities, scope, and limitations of these tools. The sessions were conducted via Zoom, with simultaneous interpretation in English and Spanish.

The academic sessions were delivered by Gabriela Alves Werb, Timur Sattarov, Sebastian Seltmann, and Georg Steinbuß, specialists from the Deutsche Bundesbank, while Dr. Gerardo Hernández del Valle, Director of Financial Market Infrastructures at CEMLA, was responsible for moderating and coordinating the two-day program. The curriculum was structured around four main areas: interpretable machine learning, causal inference, anomaly detection, and natural language processing, with particular emphasis on their relevance to economic and financial analysis.

The first day began with an introduction to the course structure and organization, followed by a review of fundamental machine learning concepts. This review established a common foundation for exploring more advanced techniques and discussing the role of machine learning models in data analysis. The program then addressed interpretable machine learning, focusing on understanding and explaining the results generated by these models. This topic is particularly relevant for central banks, where the use of analytical tools must be accompanied by approaches that make it possible to understand their outputs and assess their suitability for decision-making.

The second part of the first day was devoted to causal inference. The session introduced the main concepts, assumptions, and elements of the causal inference framework, emphasizing the importance of distinguishing between statistical relationships and cause-and-effect relationships. This distinction is fundamental to economic research and policy evaluation, as a model's ability to identify patterns or generate predictions does not necessarily imply that it can establish causal relationships. Including this topic helped place machine learning techniques within a broader methodological framework, particularly relevant to central banks' research and analytical activities.

The second day began with anomaly detection, covering its foundations and principal models. The topic was approached from the perspective of identifying observations or behaviors that deviate from usual data patterns, as well as their potential applications in finance. Anomaly detection is an important area for financial data analysis, as it can help identify unusual behavior and complement traditional procedures for monitoring and evaluating information.

Finally, the course addressed natural language processing (NLP), beginning with an introduction to its concepts and applications, followed by a discussion of generative models in central banking. This session explored the use of machine learning tools to process textual information and other unstructured data sources, as well as the opportunities offered by generative models to expand institutions' analytical capabilities. The inclusion of this topic reflected the growing importance of artificial intelligence techniques in information processing and the development of new analytical tools.

Taken together, the two sessions provided an overview of several areas of machine learning, ranging from model interpretability and causal inference to anomaly detection and natural language processing. The selection of topics combined methodological foundations with applications relevant to the financial sector and central banking, highlighting both the opportunities offered by these tools and the importance of understanding their underlying assumptions and limitations.

The theoretical component thus provided the conceptual foundation for the VII Course on Machine Learning and Central Banking and served as preparation for the in-person practical component, scheduled for September 2–4, 2026, in Panama, which focused on collaborative work and the development of use cases proposed by participants.