Hands-on course in Explainable Artificial Intelligence
We are pleased to announce that a Hands-on Course in Explainable Artificial Intelligence (XAI) will take place in the context of the 14th Iberian Gravitational Waves Meeting (IGWM2025), hosted at CIEMAT (Madrid). Participation is open to IGWM2025 attendees, as well as to members of REDONGRA or IPARCOS, regardless of whether they attend the conference.
Description
This practical course will equip participants with foundational knowledge and hands-on experience in Explainable Artificial Intelligence (XAI) techniques, focusing on their application to gravitational wave data analysis and related scientific fields. Through practical examples, participants will explore how XAI algorithms identify and highlight the most relevant features of input data that contribute to model predictions, enhancing the interpretability of AI models in physics by clarifying the relationship between input signals and model outputs. The course is especially well-suited for researchers and students interested in improving model transparency in scientific applications. All code and materials used during the session will be provided.
The course will take place in Room B on the second floor of Building 1. For practical information, please refer to the IGWM2025 web page.
Contents
- Grad-CAM Algorithm — Rationale, Strengths, and Weaknesses
Explore the application of the Grad-CAM algorithm in a classification task using a simplified version of the Gravity Spy dataset.
- LIME Algorithm — Rationale, Strengths, and Weaknesses
Explore how the LIME algorithm operates in a similar classification context, utilising a reduced Gravity Spy dataset.
- SHAP Algorithm — Rationale, Strengths, and Weaknesses
Examine the SHAP algorithm in action, applied to a classification task involving a subset of galaxies from the COSMOS2015 catalogue.
Datasets
The selected datasets were chosen based on several key criteria: their small size, facilitating hands-on demonstrations; the interpretability of their outputs; and their generality, which enhances the transferability of insights to other domains.
Required Resources and Prerequisites
Participants will need a laptop with internet access (eduroam connectivity will be provided by CIEMAT) and a valid Google account to access Google Drive and Google Colaboratory. Prior knowledge of neural networks, including experience with TensorFlow, is required, as all code examples are implemented using this framework. Familiarity with Python libraries such as NumPy, Pandas, and Matplotlib is also assumed.
Registration
This course is open exclusively to IGWM2025 attendees, as well as to members of REDONGRA or IPARCOS, regardless of whether they attend the conference. Participation is free of charge; however, registration is mandatory for all attendees. The deadline to register is 20 June.
To register, please follow this link or click on Registration in the menu on the left.
