Geometric Alignment in AI: From Optimal Transport to Efficient and Robust Algorithmic Solutions
The concept of alignment has emerged as a central theme in contemporary artificial intelligence (AI), encompassing multiple interpretations. At a high level, alignment refers to ensuring that AI systems behave in accordance with human values, norms, or ethical constraints — a perspective that underpins much of the recent discourse on AI safety.
In a more technical sense, alignment involves establishing correspondences between mathematical or computational objects, such as aligning representations from different models, data distributions, or model outputs across domains. Optimal Transport often serves as a key component in these alignment measures, offering a robust and versatile framework for complex comparisons.
In this talk, Laetitia Chapel will present contributions aimed at making Optimal Transport computationally efficient and robust, with a particular focus on structured data. By developing principled, geometry-aware, and data-efficient approaches to alignment, she addresses problems where insight, structure, and theory take precedence over brute-force computation.
Spotlight on Diverse
Career Trajectories
About the Series
Many are the young scientists wishing to pursue a career in research, but decidedly less are those knowing in which key area they should specialise. This holds especially true in the fast-paced and developing AI field, where economical, societal and scientific stakes are high; it can thus be difficult for junior researchers to reconcile current AI research with long-term goals such as a sustainable and ethical use of AI for future generations.
Furthermore, while it is relatively easy to identify the PhD degree as a first step in this sometimes daunting journey, the path forward is often unclear: what comes next after the PhD defence? How many postdoc positions is enough? How to best incorporate evolving research interests?
This new series of seminars aims to support young researchers keen to embark on an AI journey, by exploring diverse career paths within the double scope of Sustainable AI and AI for Sustainability. Sessions focus on internationally renowned speakers from diverse backgrounds whose past and/or present research strives to ensure a fair, sustainable use of AI in various fields. Following a short keynote, a discussion will follow where the guest researchers discuss different topics pertaining to their career of choice, how they came to embrace it, as well as the trials they faced along the way.