The Jülich Supercomputing Centre is organizing a training course on Introduction to Explainable Deep Learning on Supercomputers from 4 to 6 December 2023. The course will be conducted online, and the link to the streaming platform will be provided to registered participants only. Please complete registration by 20 November 2023 using the registration form.
The course aims to answer some crucial questions about Deep Learning models such as, what is the reason behind a particular prediction, what part of the input is relevant, and what changes can be made to the input to alter the output of the model. These questions are often difficult to answer as Machine Learning models are usually considered black boxes, and inspecting them is a challenging task. However, with eXplainable AI (XAI) techniques, it is possible to analyze models and reveal human-interpretable explanations of their functions. This course focuses on the concepts of XAI, particularly in Deep Learning, and provides a comprehensive understanding of its techniques.
In this course, you will gain a solid foundational understanding of XAI, with a primary focus on how XAI methodologies can help identify hidden biases in datasets and uncover valuable insights. The course begins with a broad introduction to XAI, laying the groundwork for a deeper exploration of the latest model-agnostic interpretation techniques. As the course progresses, the focus will shift to model-specific post-hoc interpretation methods. Through hands-on training, you will learn how to interpret machine learning algorithms and unravel the complexities of deep neural networks, including convolutional neural networks and transformers. You will also become proficient in applying these techniques to various data formats, such as tabular data, images, and 1D data.
During the course, participants will have the chance to not only gain theoretical knowledge but also put it into practice through hands-on sessions. This is an excellent opportunity to improve your skills in XAI and learn how to navigate the complex field of AI interpretability effectively.
