Onno Zoeter

Phone : +33 (0)4 76 61 50 71

Research Interests

Keywords: (Bayesian) Machine Learning, Mechanism Design, Approximate Inference Demand Management, Transportation.

I am interested in constructing systems that take optimal actions in an environment that is driven by random events and intentional actions of self-interested agents. In particular I am intrigued by the following observation: Learning models of human behaviour and using them to rank, filter, assign, etc. are fundamentally different problems from their analogues in say biology and medicine: humans will be aware of their impact on the system and adapt their behaviour.

When applying machine learning approaches to human data sources we need to take into account incentives. Recent results from the fields of mechanism design, Bayesian statistics, and optimal control allow us to do so.

There are a surprising number of important applications in the recent e-commerce and big data trends that have the property that a central system needs to learn and coordinate among strategic agents. Examples of applications I have worked on include on-line advertisement selection, outsourcing, worker quality tracking and selection, and demand management for transportation. Since 2010 I am leading a team of researchers that works on this last topic and this takes most of my time.

Demand management, the combination of demand based pricing and smart information provisioning systems, has in recent years found several first-of-its-kind applications, with cordon pricing in London, time differentiated tolling in Sweden and Singapore, and demand based parking rates in LA. The principles for these projects go back at least to papers William Vickrey published in the 1950's. He made the argument that public utilities such as highways, parking spaces, and public transport, should not be accessible for free, even though they are paid for by general tax money. Giving free access leads to congestion and general inefficient use: people that with little inconvenience can avoid peak hours and peak locations, have not enough incentives to do so. People familiar with Vickrey's second price auction will find it interesting to know that the development of the two ideas occurred pretty much at the same time, and both in the second price auction and in demand management the fundamental principle is to let strategic agents pay the externality they impose on society to ensure the most efficient use of scarce resources.

Our team develops Vickrey's ideas further and makes them ready for a practical deployment. We design and implement demand based pricing algorithms, study patterns in data obtained from on-street sensors and develop methods to communicate the changes in rates. Our work is in active use in Los Angeles for on-street parking (www.laexpresspark.org ). The projects has been mentioned in the MIT Technology Review when it selected Xerox among the 50 most disruptive companies in 2013 (www2.technologyreview.com/tr50/2013/ ) and was awarded the International Transport Forum's 2014 Promising Innovation in Transport Award (2014.internationaltransportforum.org/awards ).

Selected Publications from the Period Before I Joined Xerox Research

  • Onno Zoeter and Chris Dance  (2008).
    Learning optimally from self-interested data sources in on-line ad auctions.
    Abstract at BSCIW 2008.
  • Onno Zoeter (2008).
    On a form of advertiser cheating in sponsored search and a dynamic-VCG solution.
    In: Proceedings of TROA 2008.
  • Nick Craswell , Onno Zoeter, Michael Taylor, and Bill Ramsey (2008).
    An experimental comparison of click position-bias models.
    In: Proceedings of WSDM 2008.
  • Onno Zoeter, Nick Craswell , Michael Taylor, John Guiver, and Ed Snelson  (2007).
    A decision theoretic framework for implicit relevance feedback.
    NIPS 2007 Workshop: Machine learning for web search.
  • Onno Zoeter (2007).
    Bayesian generalized linear models in a terabyte world.
    In: Proceedings IEEE ISPA 2007
  • Onno Zoeter, Alexander Ypma, and Tom Heskes , (2006).
    Deterministic and stochastic Gaussian particle smoothing.
    In: Proceedings NSSPW 2006
  • Onno Zoeter, and Tom Heskes , (2006).
    Deterministic approximate inference techniques for conditionally Gaussian state space models.
    Statistics and Computing 16:279-292 [journal site ]
  • Onno Zoeter, and Tom Heskes , (2005).
    Changepoint problems in linear dynamical systems.
    Journal for Machine Learning Research (JMLR) 6: 1999-2026. [journal site ]
    Slides from the Pascal workshop Optimization and Inference in Machine Learning and Physics, Lavin 2005.
  • Tom Heskes , Manfred Opper , Wim Wiegerinck , Ole Winther , Onno Zoeter (2005).
    Approximate inference techniques with expectation constraints.
    In: Journal of Statistical Mechanics: Theory and Experiment, 2005:P11015 [journal site ] [pdf ]
  • Onno Zoeter (2005).
    Monitoring non-linear and switching dynamical systems.
    Ph.D Thesis, Radboud University Nijmegen
  • Onno Zoeter, and Tom Heskes , (2005)
    Gaussian Quadrature Based Expectation Propagation
    In: Proceedings AISTATS 2005, eds. Z. Ghahramani and R. Cowell.
  • Onno Zoeter, Alexander Ypma , and Tom Heskes , (2004).
    Improved unscented Kalman smoothing for stock volatility estimation..
    In: Proceedings of the IEEE workshop on Machine Learning for Signal Processing, eds. A. Barros, J. Principe, J. Larsen, T. Adali, and S. Douglas.
  • Tom Heskes , Onno Zoeter, and Wim Wiegerinck  (2004).
    Approximate Expectation Maximization.
    In: Proceedings NIPS 16. [ps.gz ][pdf ]
  • Onno Zoeter and Tom Heskes  (2003).
    Hierarchical visualization of time-series data using switching linear dynamical systems.
    IEEE Transactions on Pattern Analysis and Machine Intelligence Vol. 25, No. 10, October 2003, pp. 1202-1215.
  •  Onno Zoeter and Tom Heskes  (2003).
     Multi-scale switching linear dynamical systems.
     In: Proceedings ICANN/ICONIP 2003.
  • Tom Heskes  and Onno Zoeter (2003).
    Generalized belief propagation for approximate inference in hybrid Bayesian networks.
    In: Proceedings AISTATS 2003, eds. C. Bishop and B. Frey.
  • Tom Heskes  and Onno Zoeter (2002).
    Expectation propagation for approximate inference in dynamic Bayesian networks.
    In: Proceedings UAI-2002, eds. A. Darwiche and N. Friedman, pp. 216-223. [extended technical report ]