Jonas Geiping
Jonas Geiping researches safety, security, privacy and efficiency in machine learning. His work with collaborators showed that shared gradients can reveal training images in federated learning. He earned his computer science doctorate at Siegen in 2021 and researched at Maryland before moving to Tübingen in 2023. With Tom Goldstein, he studied training a masked language model from scratch on one GPU in one day. He leads a joint ELLIS and Max Planck research group focused on safety and efficiency.
- Works
- 112
- Citations
- 1,628
- h-index
- 17
Publication numbers from OpenAlex, as of 14 October 2026.
Timeline
-
2020
Examine privacy in shared updates
Geiping and collaborators studied reconstructing training images from parameter gradients shared during federated learning. Their paper demonstrated image recovery in several evaluated settings, showing that keeping raw data on a device does not by itself establish privacy.
-
2021
Doctoral research at Siegen
He earned a computer science doctorate at Siegen in 2021 under Michael Möller. He also earned mathematics degrees at Münster and later did postdoctoral work with Tom Goldstein at Maryland.
-
2022
Train with a strict computing budget
Geiping and Goldstein’s Cramming study asks how far masked language-model training can go on one consumer GPU in one day. It changes the training pipeline and evaluates downstream performance.
-
2023 onward
Safety and efficiency in Tübingen
He moved to Tübingen in 2023 and leads a joint group at ELLIS and Max Planck. The group investigates how intelligent systems can be constructed to improve safety, alongside research on efficiency and language-model behavior.
Sources and credits
Photo: ELLIS profile; creator not stated. License: Not stated by source. Source: ellis.eu. Cropped to a square and resized. Ask us to remove or replace this photo.