Researcher

Celestine Mendler-Dünner

Also known as Celestine Mendler-Dunner

Celestine Mendler-Dünner trained at ETH Zurich in collaboration with IBM Research and co-led work on Snap ML. The team published a 2018 architecture paper on Snap ML. In 2020, she co-authored Performative Prediction, which studies how deployed predictions can change the outcomes and data that later models observe. Her later work includes algorithmic collective action and tests of language-model survey responses. She held roles at Berkeley and Tübingen before taking an ETH professorship.

Works
31
Citations
159
h-index
7

Publication numbers from OpenAlex, as of 14 October 2026.

Photo of Celestine Mendler-Dünner

Timeline

  1. Doctoral work and 2018

    Learning algorithms and computing systems

    Mendler-Dünner conducted doctoral research at ETH with IBM Research Zurich. She co-led development of Snap ML and links this work to a 2018 architecture paper. The project studies how learning algorithms can use available computing systems and resources efficiently.

  2. 2020

    Predictions that change their target

    She co-authored Performative Prediction with Juan Perdomo, Tijana Zrnic, and Moritz Hardt at ICML 2020. The paper studies decisions based on predictions that change the outcomes being predicted. It gives a risk minimisation framework and a definition of performative stability.

  3. 2024

    Testing language-model survey responses

    She coauthored Questioning the Survey Responses of Large Language Models at NeurIPS 2024. The paper examines whether model survey responses can supply valid information about human populations. It also provides Folktexts resources for testing predictions against survey data.

  4. As of October 2026

    Research on learning and society

    She conducted Berkeley postdoctoral work and later led research at Tübingen before becoming a professor at ETH. Her research includes digital economic power and algorithmic collective action. These topics examine feedback and incentives around deployed AI.

Sources and credits

Photo: Tübingen AI Center / Elia Schmid. License: Not stated by source. Source: celestine.ai. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. Researcher career and project record celestine.ai. Checked 13 October 2026.
  2. Performative Prediction proceedings.mlr.press. Checked 13 October 2026.