Tosca Lechner

I am a Postdocotoral Fellow at the Vector Insitute, supervised by Professor Dan Roy and Professor Sivan Sabato. Before working at the Vector Institute I did my PhD at the University of Waterloo working with Professor Shai Ben-David. I did my masters at the University of Tübingen in Cognitive Science. I wrote my master thesis on "Domain Adaptation and Causal Assumptions" at the Max-Planck Institute for Intelligent Systems under the supervision of Professor Ruth Urner. Prior to that, I did my bachelor's degree in Mathematics at Ludwig-Maximilians-Universität, Munich.

I am interested in facilitating a better understanding of Machine Learning models and their limitations by means of a mathematical analysis. I aim to find intuitive assumptions that are meaningful in practice and lead to formal guarantees in scenarios in which common statistical assumptions break down, such as transfer learning. I believe that by developing tools that make the limitations of a model more explicit, it becomes easier to assess the trustworthiness of a model's prediction. My current research interests include adversarially and strategically robust learning, distribution learning, transfer learning and algorithmic fairness.

Publications

• On the Learnability of Distribution Classes with Adaptive Adversaries. Tosca Lechner, Alex Bie, Gautam Kamath ICML 2025. July 2025.
• On the computability of robust pac learning . Pascale Gourdeau, Tosca Lechner, Ruth Urner COLT 2024. July 2024.
• Inherent limitations of dimensions for characterizing learnability of distribution classes . Tosca Lechner, Shai Ben-David COLT 2024. July 2024.
• Distribution Learning and Robustness. Shai Ben-David, Alex Bie, Gautam Kamath and Tosca Lechner NeurIPS 2023. December 2023.
• Adversarially Robust Learning with Uncertain Perturbation Sets. Tosca Lechner, Vinayak Pathak and Ruth Urner NeurIPS 2023. December 2023.
• Strategic Classification with Unknown User Manipulations. Tosca Lechner, Ruth Urner and Shai Ben-David ICML 2023. July 2023.
• Inherent Limitations of Multi-Task Fair Representations. Tosca Lechner and Shai Ben-David CoLLAs 2022 (extended arXiv version with Nivasini Ananthakrishnan and Sushant Agarwal). August 2022.
• Learning Losses for Strategic Classification. Tosca Lechner and Ruth Urner AAAI 2022. February 2022.
• Identifying Regions of Trusted Predictions. Nivasini Ananthakrishnan, Shai Ben-David, Tosca Lechner and Ruth Urner UAI 2021. July 2021.
• On Learnability with Computable Learners . Sushant Agarwal, Nivasini Ananthakrishnan, Shai Ben-David, Tosca Lechner and Ruth Urner ALT 2020. February 2020.
• Domain Adaptation under Causal Assumptions. Tosca Lechner Master's Thesis. October 2018.

Design borrowed from vikrams.