SML group at ETH

Department of Computer Science, ETH Zurich.

Welcome to the website of the Statistical Machine Learning group at ETH Zurich!

We are a group of curious minds from different parts of the world who study exciting questions in the intersection of statistics and machine learning. On a high level, we like to develop theoretical understanding for methodological advancements and vice versa.

Most projects in our group revolve around trustworthy modern machine learning including robust inference against attacks during training and test time, generalization of overparameterized models, as well as questions arising around privacy preservation of sensitive data. Please have a look at our recent papers to get a better sense for our research interests. For example, we are currently exploring the fundamental limits, methodologies and trade-offs between different notions of trustworthiness such as (train and test-time) robustness, privacy and fairness.

news

Jan 20, 2024 Four of our recent submissions got accepted at the 2024 editions of AISTATS, SaTML and ALT. On privacy: Effective semi-supervised semi-private learning, private data synthesis for sparse queries; On causal inference: quantifying hidden confounding; On overparameterization and interpolation: max-l1-margin classifiers
Dec 10, 2023 Thanks to the crowd who showed up at our Neurips Tutorial on overparameterization! You can still watch the video and find more information on the tutorial website. Also gave a talk at ICSDS Lisbon on a line of work we just started on quantifying hidden confounding with additional datasets.
Nov 15, 2023 We’re delighted and grateful to have received the Starting Grant! We invite postdoctoral candidates with an excellent background and interest in mathematical foundations of trustworthy (statistical) machine learning to contact Fanny.
Sep 20, 2023 Later this year at NeurIPS 23 in New Orleans, Fanny will present the tutorial “Reconsidering Overfitting in the Age of Overparameterized Models” together with Vidya Muthukumar and Spencer Frei. Further, our paper “Can semi-supervised learning use all the data effectively? A lower bound perspective” was accepted as a spotlight!
Jul 30, 2023 In May, we organized the TrustML-Un(Limited) at ICLR 23 in Kigali, Rwanda and presented the paper on provable inefficiency of uncertainty sampling at ICML 23 in Hawaii, in July.