About me
Machine learning and artificial intelligence for sovereignty and democratic values.
My work sits where statistical science, machine learning and engineering meet practice. For years I have worked to bridge the gap between industry and academia, pushing the boundaries of what machine learning (ML) and artificial intelligence (AI) can do and how they can be applied to solve real problems. I believe in the positive impact that ML and AI can have on society, but I am wary of the dangers of unchecked agentic warfare. That concern is part of why I have chosen to focus on Sovereign AI and European defence. I would prefer this were not necessary; the reality is that Europe must become self-sufficient in deep tech and defence. We have the talent pool, and we must use it to defend our democratic values.
I am currently Chief AI Officer of a European sovereign AI company in stealth. Before that I was Director of Artificial Intelligence at Helsing, and earlier a Principal Applied Scientist at Amazon and Chief Scientist for Amazon SageMaker until June 2023. Since 2017 I have been an associate member of the Department of Statistics at the University of Oxford, and in 2018 I was elected Fellow in the Robust Machine Learning Programme of ELLIS, the European Laboratory for Learning and Intelligent Systems.
Background
My family taught me to do the right thing; that reading and learning matter, because knowledge, once gained, is yours to keep; and that science is the path to discovery, with mathematics its universal language. A passion for science and aesthetics took hold early, and I chose to study engineering to have a tangible impact on the world.
My Electrical Engineering degree and PhD in Applied Sciences are from UCLouvain, earned in 2001 and 2005, where I was a member of the Machine Learning Group and the Crypto Group. I worked on the European projects OPTIVIP, developing neural networks embedded in a visual prosthesis for the blind, and SCARD, demonstrating weaknesses of cryptographic hardware against machine learning-based side-channel attacks exploiting electromagnetic radiation.
A post-doc with John Shawe-Taylor at University College London followed. I collaborated with Manfred Opper at TU Berlin on data assimilation and approximate Bayesian inference, and took part in the PASCAL European network of excellence. Until 2015 I held an Honorary Senior Research Associate position in UCL’s Centre for Computational Statistics and Machine Learning.
In October 2009 I joined Xerox Research Centre Europe (now Naver Labs Europe), where I led the Machine Learning group, working on natural language understanding and mechanism design with applications in customer care, transportation and governmental services. One output of the team was LA Express Park, demand-based parking pricing for the City of Los Angeles.
Amazon brought me to Berlin in October 2013 to develop zero-parameter machine learning algorithms. The goal was to help software engineers from retail teams to use machine learning in their applications and products. I also established Amazon’s Limited Scientific Publication process and the Amazon Scholars programme; today Amazon hosts hundreds of academics, including Nobel laureates, and publishes at the major academic conferences.
From 2019 onwards, as a Principal Applied Scientist at AWS, I oversaw the product-related science powering Amazon SageMaker and led initiatives in automated machine learning, continual learning and responsible AI. That work laid the foundation of SageMaker Automatic Model Tuning, SageMaker Autopilot, SageMaker Clarify, SageMaker Inference Recommender, and AWS Clean Rooms. SageMaker is used by tens of thousands of customers, making more than 1.5 trillion inference requests every month. I was also an Amazon Bar Raiser and an AWS Press Spokesperson.
As the Director of AI at Helsing, I built out the AI organisation for the land domain, developing and shipping the perception capabilities deployed on Helsing’s recce-strike complex, the HX-2, one of Helsing’s flagship products, with thousands of units delivered to Ukraine. I also established Helsing’s AI Assurance and Data Sourcing teams, and was a founding member of Area 9, Helsing’s product incubator.
Research interests (over the years)
- Bayesian inference, approximate inference and variational inference
- Natural language processing and understanding
- Bayesian optimisation, hyperparameter optimisation and automated machine learning
- Responsible and trustworthy AI
- Transfer learning, parameter-efficient fine-tuning and foundation model adaptation
- Multi-modal foundation models and physical AI models