Christian Schroeder de Witt

Christian Schroeder de Witt

AI & Security Research | Strategy

University of Oxford

I am a leading researcher in foundational AI and information security. My recent works include a breakthrough result on the 25+ year old problem of perfectly secure steganography (jointly with Sam Sokota), which was featured by Scientific American , Quanta Magazine, and Bruce Schneier’s Security Blog, as well as illusory attacks, a novel form of adversarial attack on reinforcement learning agents (Spotlight at ICLR 2024). During my Ph.D., I helped establish the field of cooperative deep multi-agent reinforcement learning, resulting in popular learning algorithms such as QMIX, MACKRL, IPPO, and FACMAC, and the standard benchmark environments SMAC and Multi-Agent MuJoCo.

I am currently a postdoc with Torr Vision Group at the University of Oxford, and a former visiting researcher with Turing Award-winner Prof. Yoshua Bengio at MILA (Quebec) and postdoc with FLAIR. Previously, I completed my DPhil (Ph.D.) “Coordination and Communication in Deep Multi-Agent Reinforcement Learning” with Prof. Philip Torr (Torr Vision Group) and Prof. Shimon Whiteson (WhiRL), which won an EPSRC IAA Doctoral Impact Fund Award and is, according to my examiner Prof. Frans Oliehoek, a “standard reference in the field”.

I hold distinguished masters degrees in Physics, as well as Computer Science (both University of Oxford), during the latter of which I proved an open incompleteness theorem in categorical quantum mechanics (the completed ZX-calculus is now a mainstream tool in quantum computing).

In 2022, I was selected as a “30 under 35 rising strategist (Europe)” by Schmidt Futures International Strategy Forum and the European Council on Foreign Relations. I also received a Best Idea award from the CCAI community in 2019 for work on solar geoengineering and deep multi-agent learning.

Supervision

Please contact me if you are interested in working with me. I supervise undergraduate projects, master’s theses, and co-supervise Ph.D. projects in a wide range of topics in both deep multi-agent learning, and information security.

Some of my students and mentees include:

  • Linas Nasvytis (MSc Statistics student, now Research Fellow at Harvard University (Psychology and ML))
  • Yat Long Lo (MSc Computer Science student - winner of Tony Hoare Prize for best MSc Thesis in Computer Science, now Dyson Robot Learning Lab)
  • Khaulat Abdulhakeem (mentee, now MS Education Data Science at Stanford University)
  • Eshaan Agrawal (mentee and collaborator, now ORISE Fellow at the Department of Energy)
Interests
  • Multi-Agent Learning
  • Information Security
  • Reinforcement Learning
  • Agent-Based Modeling
  • Cooperative AI
Education
  • DPhil (PhD) in Engineering Science, Jan 2017 - Nov 2021

    University of Oxford, St. Catherine's College

  • MSc Computer Science - Distinction, Oct 2012 - Sep 2013

    University of Oxford, Kellogg College

  • MPhys (Physics) - First, Oct 2008 - Jul 2012

    University of Oxford, Exeter College

Recent News

All news»

[16/01/24] Many congratulations to DPhil student Tim Franzmeyer for winning a Spotlight at ICLR'2024 for our project Illusory Attacks!

[21/12/23] My Master’s student Linas Nasvytis’ work on Out-of-Distribution Dynamics Detection for Reinforcement Learning has been accepted for an Oral Presentation at AAMAS 2024. Congratulations, Linas!

[1/09/23] My work on Perfectly Secure Steganography (with Sam Sokota) is featured on Bruce Schneier’s Security Blog!

[1/09/23] My work on Perfectly Secure Steganography (with Sam Sokota) is featured on Bruce Schneier’s Security Blog!

[31/08/23] My work on Perfectly Secure Steganography (with Sam Sokota) is featured in Scientific American (both print and online)!

Experience

 
 
 
 
 
University of Oxford
Postdoctoral Research Assistant
Jan 2022 – Present Oxford, United Kingdom
Since 2024 working with Torr Vision Group (TVG). Funded by Prof. Phil Torr’s Turing AI World-Leading Researcher Fellowship. Formerly with FLAIR (Prof. Jakob Foerster). Chair of AI4ABM community.
 
 
 
 
 
Schmidt Futures
International Strategy Forum | Fellow
Jun 2022 – Present NY, remote
Selected to work with the European Council on Foreign Affairs as one of Europe’s top 30 under 35 rising strategists in technology and geopolitics. Nominated by the Vice-Chancellor of the University of Oxford as “one of the most exemplar individuals [she has] encountered in the past ten years.”
 
 
 
 
 
MILA (Quebec)
Visiting Researcher
Aug 2022 – Present Montreal, remote
Working with Turing-Award winner Prof. Yoshua Bengio.
 
 
 
 
 
MenschDanke Group
Head of Engineering
Mar 2014 – Sep 2014 Berlin (DE)
Interim management of a team of 4 in-house developers. Full-stack development (LAMP, MEAN) and agile project management (Scrum) at Germany’s then third-largest e-Commerce (deals) venture. Negotiated long-term hardware contracts and supervised live product migrations.

Honors, Awards & Grants

As an official Co-Editor, I helped with Prof. Jakob Foerster’s ERC Starter Grant.
I was awarded an EPSRC IAA Doctoral Impact Fund Award (£30k) for my DPhil thesis.
I reached the global finals (stage 4/4) of this ultra-competitive fellowship with a NASA-backed proposal on mesospheric geoengineering research.
De-Facto PI of a series of grant arising out of my ongoing personal long-term relationship with Armasuisse Science+Technology (Zurich, CH). Attracted > £190k since 2021.
Awarded for climate science research.
In Germany the top 0.5% of high school graduates get selected for funding by the German Academic Foundation (also known as German Merit Foundation).
Awarded a special prize at Brandon Festival of the Arts, and two first prizes.

Publications (Selected)

Quickly discover relevant content by filtering publications.
Communicating via Markov Decision Processes
We propose a perfectly-secure steganography algorithm for arbitrary covertext distributions.
Communicating via Markov Decision Processes
Amortized Rejection Sampling in Universal Probabilistic Programming
In this paper we develop a new and efficient amortized importance sampling estimator for rejection sampling.
Amortized Rejection Sampling in Universal Probabilistic Programming
Discovered Policy Optimisation
In this paper we explore the Mirror Learning space by meta-learning a “drift” function.
Discovered Policy Optimisation
Rainbench: Towards Data-Driven Global Precipitation Forecasting from Satellite Imagery
We introduce RainBench, a new multi-modal benchmark dataset for data-driven precipitation forecasting.
Rainbench: Towards Data-Driven Global Precipitation Forecasting from Satellite Imagery
FACMAC: Factored Multi-Agent Centralised Policy Gradients
We propose multi-agent common knowledge reinforcement learning (MACKRL).
FACMAC: Factored Multi-Agent Centralised Policy Gradients
Multi-Agent Common Knowledge Reinforcement Learning
We propose multi-agent common knowledge reinforcement learning (MACKRL).
Multi-Agent Common Knowledge Reinforcement Learning
The Starcraft Multi-Agent Challenge
We propose the StarCraft Multi-Agent Challenge (SMAC) to measure real progress in MARL.
The Starcraft Multi-Agent Challenge

Outreach

Multi-Agent Security Community (MASEC)
I co-founded the ABM community, and, as its Chair, am the main organiser of a NeurIPS 2023 workshop. We have received sponsorship from GovAI.
Multi-Agent Security Community (MASEC)
AI for Agent-Based Modelling Community (AI4ABM)
I co-founded the ABM community, and, as its Chair, have been the main organiser of workshops at ICML 2022 (sponsored by Improbable), and ICLR 2023 (sponsored by JP Morgan). Agent-Based Modelling (ABM) has seen increasing interest across various disciplines, ranging from economics and epidemiology to cybersecurity, social sciences, and climate policy.
AI for Agent-Based Modelling Community (AI4ABM)

Press

AI Could Smuggle Secret Messages in Memes
A new technique for sending hidden messages is mathematically proven to escape detection
AI Could Smuggle Secret Messages in Memes
AI-Generated Steganography
New research suggests that AIs can produce perfectly secure steganographic images
AI-Generated Steganography
Secret Messages Can Hide in AI-Generated Media
In steganography, an ordinary message masks the presence of a secret communication. Humans can never do it perfectly, but a new study shows it’s possible for machines.
Secret Messages Can Hide in AI-Generated Media
New Breakthrough enables Perfectly Secure Digital Communications
A team led by University of Oxford researchers has achieved a breakthrough in secure communications by developing an algorithm that conceals sensitive information so effectively that it is impossible to detect that anything has been hidden.
New Breakthrough enables Perfectly Secure Digital Communications
“Perfectly secure” algorithm could aid spread of free speech
A new algorithm has solid implications for information security, data compression and storage, but the real benefits could be seen among vulnerable groups
“Perfectly secure” algorithm could aid spread of free speech
Steganography algorithms enable ‘perfectly secure’ information
An algorithm has been developed that conceals sensitive information so effectively that it is impossible detect that anything has been hidden.
Steganography algorithms enable ‘perfectly secure’ information
Climate Change: the Case for Artificial Intelligence
A recent Intergovernmental Panel on Climate Change (IPCC) report has made it very clear that drastic, immediate cuts to greenhouse gas emissions are needed to limit global warming to 1.5° C. With the absence of a technological silver bullet, this requires rapid changes at unprecedented scale across all sectors of the global economy. Just as the climate clock is ticking, technological breakthroughs in machine learning algorithms (ML) and robotic control have turned artificial intelligence (AI) into a powerful new agent of change.
Climate Change: the Case for Artificial Intelligence

Blog

*
The Sweeping Powers of Model-Free Opponent Shaping
Opponent shaping is a powerful technique that can be used to induce cooperation, but also to extort others. New, model-free techniques exemplify this approach and provide lessons for the development of Cooperative AI.
The Sweeping Powers of Model-Free Opponent Shaping

Consulting

Contact me here for 1:1 consulting on topics related to my past industry research experiences and/or my current research in academia.

    Industry Collaboration

    Contact me here if you would like me to complete an end-to-end industry research project. For reference, projects I take on typically lead to patents and/or papers.

      Speaker/Moderator Request

      Contact me here if you would like me to speak at or moderate an event such as a panel discussion, podcast episode, or interview.

        Other

        Contact me by email for a purpose other than the ones listed above.