Biography

I’m a Senior Research Scientist at Google DeepMind in the science team. Before that I was a Research Scientist at Elekta and pursued a PhD in Applied Mathematics under the supervision of Ozan Öktem.

At DeepMind I work on scientific machine learning, especially focusing on inverse problems and the intersection between model-driven and data-driven methods. Among other things I’ve worked on AlphaFold, recongnized as a solution to the protein folding problem at CASP14.

May 2020: New pre-print out: Accelerated Forward-Backward Optimization using Deep Learning.

Nov 2020: We have announced the new version of AlphaFold at CASP14.

Jun 2019: I’m joining the Science team at DeepMind.

Feb 2019: Giving an invited talk at BASP: Deep Posterior Sampling.

Jan 2019: Organising DLIP and giving a talk on Deep Bayesian Inversion.

Dec 2018: I’ll be travelling the US east coast, giving talks at Johns Hopkins on Tuesday Dec 11, at New York University Wednesday Dec 12 and at Flatiron Institute Thursday Dec 13.

Dec 2018: Giving an oral on MedNeurIPS on Task adapted reconstruction for inverse problems Saturday, Dec 8 at 11:00.

Nov 2018: New preprint: Deep Bayesian Inversion.

Nov 2018: Visiting Matthias Ehrhardt at the University of Bath to give a seminar on Deep Learning for Image Reconstruction Tuesday, Nov 20 at 13:15.

Oct 2018: Our extended abstract for Task adapted reconstruction for inverse problems has been accepted at Medical Imaging meets NIPS.

Oct 2018: Extended visit to Prof. Carola Schönliebs group in Cambridge and the Alan Turing Institute from 29 Oct to 23 Nov.

Oct 2018: Presenting Deep Learning for Image Reconstruction at the Chinese Academy of Sciences.

Sept 2018: Organizing the international workshop Deep Learning and Inverse Problems 21-25 Jan 2019 in Stockholm.

Sept 2018: Banach Wasserstein GAN has been accepted for publication at NIPS .

Sept 2018: New preprint: Task adapted reconstruction for inverse problems.

ODL

Operator Discretization Library (ODL) is a Python library that enables research in inverse problems on realistic or real data.

Minimal ML implementations

Some minimalistic implementations of generative models:

• VAE
• WGAN-GP