The ARMEDIA team is pleased to invite you to its research seminar on:

“Towards Human–Machine Alignment in Explainable AI: From Machine-Centered Attributions to the Information Conveyed by Explanations”

by Felipe Torres Figueroa, on June 25th at 2:00 PM, in room A005-006 on the TSP Évry campus and online at 

Séminaire ARMEDIA | Réunion-Joindre | Microsoft Teams

Abstract

With the advent of deep learning over the past decade, numerous Artificial Intelligence (AI) methods and tools have become pervasive in society. These advances have enabled the automation of tasks such as autonomous driving, medical diagnosis and tool design, while also enhancing our understanding of the visual world. Nevertheless, as these technologies continue to evolve rapidly, it is essential to consider key aspects to mitigate their potential negative impacts on society.

In this context, the field of Explainable AI (XAI) seeks to address questions related to the functioning of these methods. It aims to promote a deeper understanding of AI systems, facilitate model refinement for researchers, and bridge the gap in understanding for non-expert users.

In this talk, I will briefly introduce the main concepts of XAI before focusing on two broader discussions. First, I will present Opti-CAM, an attribution method that generates saliency maps by maximizing the predicted probability of a target class. Second, building on this, I will discuss challenges related to human–machine alignment in interpretability studies, from both academic perspectives and the viewpoint of non-expert users.

Finally, I will introduce a proposed study on this alignment, which aims to investigate how different types of information conveyed by explanations are understood by both humans and AI systems.

Speaker short biography

Felipe Torres is a postdoctoral researcher at Institut Mines-Télécom Business School, where he is currently working on the ENFIELD project, which aims to develop explainable AI (XAI) technologies to better understand AI models and bridge the gap between human and machine understanding.

Originally trained as a biomedical engineer, Felipe earned his PhD from École Centrale de Marseille, where he began his research in XAI. He has published multidisciplinary work at the intersection of computer vision and engineering applications, including a tool for measuring wound size in microscopy images, as well as methods such as Opti-CAM for generating explanations in image recognition models.

We look forward to your participation.