Demystifying Transformers: foundations, architecture and interpretability

Seminari

Title: Demystifying Transformers: foundations, architecture and interpretability.

Dates: April 16th, 2026.Time: 9:30-11:30 (break) 12:00-14:00.Hours: 4 h.

Instructor: Laia Tarrés Benet (UPF)

Languages: Spanish/English

Modality: The seminar will be in person (no online modality) 

Location: The seminar will be held at the Escola d’Enginyeria. Room to be determined.

Registration: The seminar is FREE but registration is required. 

Please use https://forms.office.com/e/Gs2uuT1wGc to register.

Summary: Deep learning has become established over the last decade as one of the key technologies in artificial intelligence, driving cutting-edge advances in computer vision, natural language processing, robotics and many other areas. This seminar is designed for computer science students seeking to confidently engage with transformer-based systems in research contexts. The seminar will provide an overview of deep learning methods and will break down the core principles behind transformer architectures, from attention mechanisms and model design to training strategies, before moving into the increasingly important topic of interpretability. Participants will gain insight into how these models work internally, how to analyze their behavior and how to apply them in research scenarios. The seminar will explore how transformer architectures can be leveraged across diverse research domains, with a focus on adapting pre-trained models to specific tasks through techniques such as fine-tuning and transfer learning.