Large AI Models (LAIMs), of which large language models are the most prominent recent
example, showcase some impressive performance. However they have been empirically …

The geometric median, an instrumental component of the secure machine learning toolbox,
is known to be effective when robustly aggregating models (or gradients), gathered from …

In supervised learning-for instance in image classification-modern massive datasets are
commonly labeled by a crowd of workers. The obtained labels in this crowdsourcing setting …

Many applications, such as content moderation and recommendation, require reviewing and
scoring a large number of alternatives. Doing so robustly is however very challenging …

In this article, we argue that YouTube's algorithm should be programmed to make a modest
but significant percentage (eg 2%) of recommendations for the climate. Just as a librarian …

Many applications, eg in content recommendation, sports, or recruitment, leverage the
comparisons of alternatives to score those alternatives. The classical Bradley-Terry model …

When queried, humans do not always give their utmost attention to provide their best
possible answer. In particular, they may give different answers to the same question …

Recommendation algorithms play an increasingly central role in our information ecosystem.
Yet, so far, they are mostly designed, parameterized and updated unilaterally by private …

Phishing has been one of the most frequent cyber threats in the recent decade, prompting
an increase in anti-phishing research and the development of numerous solutions for …

In supervised learning--for instance in image classification--modern massive datasets are
commonly labelled by a crowd of workers. The obtained labels in this crowdsourcing setting …