Building foundational audio&music models and their applications

June 2023 - present: Science of Scale @AWS-Lablets -- "Learning to Scale Team" to improve the training of large foundational modelsNov. 2021 - June 2023: developing methods to assess and improve robustness and fairness of computer vision models

Group: Learning and Adaptive Systems, lead by: Prof. Andreas KrauseI plan to continue my work on adversarial examples, as well as to explore the links between causality, adversarial learning and reinforcement learning.

Advisor: Bernhard Schölkopf- Understanding adversarial examples- Theory of Generative Adversarial Networks (GANs)- Kernel based sample tests, kernel based metrics on probability distributions- Development of causality based algorithms for exoplanet detection- Population structure correction in GWAS- Subphenotype discovery in clinical diseases

In recent years, state-of-the-art convolutional network classifiers have been found utterly vulnerable to adversarial examples: their performance can be driven down by almost invisible -but targeted- perturbations of the input images.During my internship, I worked together with Yann Ollivier, Léon Bottou and David Lopez-Paz to help understanding the origin of this adversarial vulnerability.

Advisor: Jean-Philippe VertStatistics and machine learning methods applied both to the study of phenotypic and genotypic properties of different yeast cell strains and to the modeling of the genetic recombination events (crossovers and non-crossovers) during meiosis

Development of a pricing program to evaluate the cost of standard depots.Deputy Site Manager for the building of a depot in Lauenburg (near Hamburg).