Jonas Wulff

Also known as Jonas Sebastian Wulff

Jonas Wulff is a computer vision and machine learning researcher who helped turn the open-source animated film Sintel into MPI Sintel, a benchmark for optical flow. As a PhD student in Michael Black's group at the Max Planck Institute for Intelligent Systems, he fixed how Blender rendered motion for the 2012 dataset, then developed the PCA-Flow and MR-Flow methods. From 2018 to 2021, he worked with Antonio Torralba at MIT CSAIL on the representations inside deep networks, including a 2019 study of what GANs cannot generate. The Sintel paper received the ECCV Koenderink Prize in 2022, and at Xyla he coauthored the 2023 CHIL best paper on clinical language models. Since 2023, he is ML lead at OpenEvidence, the medical AI company formerly called Xyla, where he is a coinventor on a medical question-answering patent granted in August 2026.

Now at: OpenEvidence
Badges: Bridge

Timeline

2005–2011

Computer engineering at RWTH Aachen

He studies for a computer engineering diploma at RWTH Aachen University from 2005 to 2011. With Thomas Stehle and Til Aach, he coauthors medical imaging papers such as a 2009 ISBI paper on a motion-compensated video filter for denoising fluorescence endoscopy.
2011–2012

Building the MPI Sintel benchmark

He joins Michael Black's group at the Max Planck Institute for Intelligent Systems in Tübingen as a PhD student in 2011 and fixes how Blender renders optical flow from the open-source film Sintel. The ECCV 2012 paper "A Naturalistic Open Source Movie for Optical Flow Evaluation," with Daniel Butler, Garrett Stanley and Black, presents the resulting MPI Sintel dataset.
2015–2018

PCA-Flow, MR-Flow and a PhD

With Black, he presents PCA-Flow at CVPR 2015, which approximates optical flow as a weighted sum of basis flow fields learned from four Hollywood movies. In 2017, he presents MR-Flow at CVPR with Laura Sevilla-Lara and Black, reaching state-of-the-art results on MPI-Sintel and KITTI-2015. He defends his University of Tübingen dissertation, "Model-based Optical Flow: Layers, Learning, and Geometry," in April 2018, with Black as reviewer.
2018–2021

Postdoc with Antonio Torralba

From 2018 to 2021, he is a postdoctoral researcher with Antonio Torralba at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), studying the representations inside deep neural networks. He coauthors "Seeing What a GAN Cannot Generate" with David Bau and others, an ICCV 2019 oral on GAN mode collapse. In 2021, "Learning to See by Looking at Noise" trains visual representation learners on images made by simple noise processes.
2021–2023

Clinical language models at Xyla

He becomes a research scientist at Xyla in September 2021 and works on natural language processing. In 2023, he coauthors "Do We Still Need Clinical Language Models?" with Eric Lehman and others, which finds that small specialized clinical models outperform in-context learning with general models and wins Best Paper at the 2023 Conference on Health, Inference, and Learning (CHIL).
2022

The Koenderink Prize

In 2022, he shares the ECCV Koenderink Prize, which recognizes work that has stood the test of time, with Daniel Butler, Garrett Stanley and Michael Black for their 2012 Sintel paper. By 2022, 375 methods are publicly ranked on the Sintel evaluation server.
Since 2023

ML lead at OpenEvidence

He becomes ML lead at OpenEvidence, the medical AI company formerly called Xyla, in 2023. In July 2023, OpenEvidence counts him among its key scientists from Harvard and MIT AI labs.
December 2024–August 2026

A medical question-answering patent

He is a coinventor on OpenEvidence's medical question-answering patent, filed in December 2024 and granted in August 2026. The system answers medical queries from a corpus of documents and shows a separate answer based only on clinical practice guidelines.