Christian Schroeder de Witt

Christian Schroeder de Witt

DPhil Student


I am a 2nd-year PhD student conducting fundamental algorithmic research in artificial intelligence (AI), machine learning, computer vision and autonomous control (particularly, UAV control). My supervision is jointly between Prof. Shimon Whiteson and Prof. Philip Torr (Torr Vision Group). In this capacity, I am funded by Free The Drones (Innovationsfonden Denmark) and Microsoft UK.

On the algorithmic side, I am working on deep multi-agent reinforcement learning and autonomous control. In particular, I am interested in central learning with decentralized execution, multi-agent exploration and large-scale inter-agent coordination and interaction. Recent publications in this space include QMIX [1] and MACKRL [2]. Benchmarks I am particularly interested in include StarCraftI/II™, large-scale kinematic models and sensor networks. For a full description of my work spanning deep reinforcement learning and autonomous control, please refer to my Torr Vision Group profile.

Other interests

Before returning to Oxford for a PhD, I worked for a year as a research associate for theoretical machine learning (convex optimisation, statistical learning theory) at Humboldt University, Berlin [3]. I also hold two master’s degrees from the University of Oxford: One in Physics (MPhys, First and University Thesis Prize, thesis with Prof Ard Louis [6]), and one in Computer Science (MSc, Distinction, thesis with Prof. Bob Coecke [5]). During my time as a full-time quantum physicist, I proved an important incompleteness theorem for ZX-calculus, a graphical calculus for finite-dimensional quantum mechanics [4].

I also spent some time in industry, working as interim Head of Engineering of a large Berlin eCommerce company, facilitating the work of a team of eight developers. In summer 2017, I was ML researcher intern at Man AHL, one of the world’s largest quantitative hedgefunds.

Apart from my professional career as a researcher, I have a long history of engagement with contemporary issues, such as climate change and refuge and asylum. Forms of involvement have spanned volunteering, to co-organisation of workshops and summits all the way up to party-political activism.

I am also a trained classical concert pianist, I received a scholarship to study piano performance at Brandon University School of Music MB with Prof Megumi Masaki, in 2006, finishing top of the year in 2007 with a GPA of 4.07 before accepting my offer to study physics in Oxford.


For a full list of publications, please refer to my Google Scholar profile.

[1] Multi-Agent Common Knowledge Reinforcement Learning, Jakob Foerster*, Christian Schroeder de Witt*, Gregory Farquhar, Philip Torr, Wendelin Boehmer, Shimon Whiteson, submitted to AAAI 2019, to appear

[2] QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning, Tabish Rashid, Mikayel Samvelyan, Schroeder de Witt, Gregory Farquhar, Jakob Foerster, Shimon Whiteson, International Conference on Machine Learning, 2018,

[3]  Safe Screening for Support Vector Machines,  Julian Zimmert, Christian Schroeder de Witt, Giancarlo Kerg, and Marius Kloft,  Proceedings of the NIPS 2015 Workshop on Optimization in Machine Learning (OPT), 2015

[4]  The ZX-calculus is incomplete for quantum mechanicsChristian Schroeder de Witt and Vladimir Zamzhdiev, Presented at the 11th workshop on Quantum Physics and Logic (QPL) 2014. Published in Electronic Proceedings in Theoretical Computer Science (EPTCS),

[5]  The ZX-calculus is incomplete for non-stabilizer quantum mechanics,  Christian Schroeder de Witt, MSc Computer Science Thesis, University of Oxford (awarded Distinction), 2013

[6]  Theoretical Biology and Biological Evolution,  Christian Schroeder de Witt, MPhys Thesis, University of Oxford (awarded First and University Prize), 2012