My research is currently focused on topics in artificial intelligence
and related areas of computer science, and notably at the intersection
of machine learning, automated reasoning and optimisation;
in the past, I've also worked extensively
in bioinformatics and computer music.
My work in artificial intelligence and related areas primarily aims to leverage AI
techniques, notably combinations of machine learning and optimisation methods,
for automating the design of effective solvers for computationally challenging problems.
This helps AI experts achieve better performance with less manual effort,
and it lowers the level of expertise required for building, deploying and operating
predictable, robust and performant AI systems and tools.
Ultimately, my research has the goal of developing and facilitating the effective,
responsible use of trustworthy, human-centric AI systems that aim to complement, rather
than replace, human intelligence.
Over the past 15 years, my group has played a leading role in the automated
design and analysis of algorithms - and notably, in performance prediction, algorithm configuration, algorithm selection and construction of parallel algorithm portfolios,
which all can be seen as generalised forms of machine learning.
We have made pivotal contributions to establishing the now thriving research area of AutoML,
which deals with the automated selection and hyperparameter optimisation of machine learning algorithms.
Our work has also produced major improvements in the state of the art in solving a broad range of challenging
(NP-hard) problems, including propositional satisfiability (SAT), mixed integer programming (MIP),
the travelling salesperson problem (TSP), as well as AI planning and scheduling problems.
I also pursue some scholarly work in an area quite unrelated to computer science,
but that's another story, for another day ...
My scholarly publications, with links to electronic versions of most papers, can be found
My co-authors and I have been fortunate in having been able to produce many publications
that have been recognised with
best paper awards.
I regularly work with partners from industry, to translate advances from my fundamental research into industrial innovation,
and to draw inspiration from real-world applications. Furthermore, some of my fundamental research results are directly used
by others in impactful industrial applications. Here are some examples of both types of impact from the past 10 years:
- Working with experts at IBM, my group has leveraged our work on automated algorithm configuration for major
improvements of CPLEX, arguably the world's most widely used software for industrial optimisation.
Our work has also directly inspired similar advances in Gurobi, one of CPLEX's main competitors.
- In a recent project with a major North-American power trading company,
my former Postdoc, Lars Kotthoff, and I helped leverage automated algorithm configuration technology developed in my group for solving forecasting and pricing problems
at the core of their trading activities; as a result, forecasting horizons could be extended from 2 months to 2 years, and problems
completely out of reach prior to my work are now routinely solved in day-to-day business.
- In work with Actenum, a software company based in Vancouver (Canada), my group and I played a key role in facilitating
the use of our automated configuration and analysis techniques to develop next-generation maintenance scheduling
services for use in the oil and gas industry.
- Work on SAT modulo monotonic theories carried out jointly with my UBC colleague, Alan Hu and our PhD student, Sam Bayless,
is now used by Amazon for detecting security vulnerabilities in customer cloud networks within AWS.
- My close collaborator, Kevin Leyton-Brown (University of British Columbia, Canada), directly used our work on automated algorithm selection,
configuration and SAT solving in work with the US Federal Communication Commission (FCC), to optimise revenue in their 2016-17
bandwidth auction, resulting in a 8 billion dollar revenue for the US government.
- The ParamILS and SMAC automated algorithm configuration software developed in my group
is used by a broad range of companies from the energy, telecommunications, software development and entertainment sector,
for performance optimisation in mission-critical in-house and customer-facing applications.
- Work with my postdoc advisor, Craig Boutilier, has led to two US patents that were subsequently acquired by
CombineNet Inc, a US-based global leader in software for strategic sourcing and supply chain management (later acquired by SciQuest Inc, now JAGGAER).
I regularly present my work at conferences, workshops and other scientific meetings.
While I don't keep a complete list of these presentations (there are simply too many of them),
here are some notable examples:
If you have seen one of my presentations not listed here and would like to receive an electronic copy of the slides,
feel free to ask.
- Cooperative Competition: A New Way of Solving SAT and other NP-Hard Problems in AI and Beyond
Simons Institute for the Theory of Computing, Berkeley, CA, USA, 2021
- Learning how to solve it - faster, better and cheaper
Keynote lecture at the OR 2018 Conference in Brussels, Belgium, 2018
- Beyond Big-O: Statistical Analysis of Performance Scaling
Simons Institute for the Theory of Computing, Berkeley, CA, USA, 2016
- Machine Learning & Optimisation:
Promise and Power of Data-driven, Automated Algorithm Design
Keynote lecture at the ECAI 2018 Conference in Prague, Czech Republic, 2014
- Programming by Optimisation
University of British Columbia, Vancouver, BC, Canada 2012
My research is made possible by funding from a broad range of organisations, including
the European Commission,
NWO (the Dutch Research Council),
NSERC (the Canadian National Sciences and Research Council),
Compute Canada, BCKDF (The British Columbia Knowledge Development Fund),
NSF (the US National Science Foundation),
Leiden University, The University of British Columbia,
UBC PWIAS (the UBC Peter Wall Centre for Advanced Studies),
TUD ZIT (the TU Darmstadt Zentrum für Interdisziplinäre Technikforschung),
Studienstiftung des Deutschen Volkes,
as well as
Honda Research Institute Europe, IBM Canada, Actenum Inc., Huawei Canada and Gentel Inc.
Furthermore, many of my students are supported by scholarships.
Here are some of the projects I am currently involved in:
I am also involved in a collaboration with the European Space Agency on physics-aware automated machine learning for Earth observation
as well as in two projects on the effective and responsible use of machine learning with companies from the insurance sector.
- Software Engineering for Machine Learning
so far funded internally, but we are working on changing that - let me know if you'd like to support this line of research!
- Value and Impact through Synergy, Interaction and coOperation of Networks of AI Excellence Centres
funded by the European Commission; I am the coordinator of this project
- Trustworthy AI - Integrating Learning, Optimisation and Reasoning
funded by the European Commission; I lead the work package on AutoAI
- Making artificial intelligence human-centric
funded by the European Commission; I have a leadership role in the work package on synergies with broader European AI community
and participate in several technical work packages
- European Language Equality
funded by the European Commission; I participate on behalf of CLAIRE in order to help leverage natural language processing techniques from AI
- Hybrid Intelligence
- augmenting human intellect
10-year research programme, funded by NWO under the NWO Gravitation Programme
- Unraveling Neural Networks with Structure-Preserving Computing
funded by NWO
- Dynamic Data Analytics through automatically Constructed Machine Learning Pipelines
funded by NWO and Honda Research Institute Europe