Simon Keizer is Lecturer in the Artificial Intelligence Lab at the Vrije Universiteit Brussel (Belgium).

The main focus of his research is on dialogue management and user simulation for training and evaluating spoken dialogue systems.  In particular he is interested in using machine learning techniques in the development of interactive systems.  He has worked on several European FP7 projects, including CLASSiC and JAMES, developing such methods for multi-modal, multi-user human-robot interaction.  He was also involved in the ERC project STAC on non-cooperative interaction.  Currently, he is involved in the EPSRC project MaDrIgAL investigating cross-domain adaptation of dialogue systems via multi-dimensional dialogue modelling and transfer learning.

More general interests include computational semantics and pragmatics (dialogue act modelling), and machine learning (in particular probabilistic models such as Bayesian networks and reinforcement learning techniques, and more recently, transfer learning).  Another interest is the use of crowdsourcing technologies (such as Amazon Mechanical TurkCrowdFlowerfor evaluating spoken dialogue systems.

linkedIn Google scholar ResearchGate

Short CV

Simon Keizer studied Applied Mathematics at the University of Twente (NL).  He obtained his master's degree on the subject of knowledge graphs, a special variant of John Sowa's conceptual graphs, for representing natural language.  He then started his PhD research at the Computer Science department of the same university, in the research group Human Media Interaction.  This resulted in the successful defence of a dissertation on dealing with uncertainty in natural language dialogue using Bayesian networks. In particular, the work involved machine learning experiments on dialogue act recognition.

He continued his work on dialogue in the Department of Information and Communication Sciences at Tilburg University (NL).  The PARADIME postdoc project involved a multidimensional approach to dialogue management, in which the generation of dialogue acts is realised through several agents operating in parallel on the system's information state.  An implementation of this dialogue manager was integrated in an interactive question answering system, developed in cooperation with several other projects, specialising in question answering, syntactic & semantic analysis, speech recognition, and multimodal answer presentation.

In the Dialogue Systems Group at Cambridge University, he then worked on statistical approaches to dialogue management, in particular POMDPs (Partially-Observable Markov Decision Processes).  The focus of his work was on user modelling/simulation and dialogue systems evaluation.  He was involved in a EPSRC project as well as the EU FP7 project CLASSiC.

As a research fellow in the Interaction Lab at Heriot-Watt University in Edinburgh, Simon has worked on several projects, including JAMES (human robot interaction) and STAC (strategic conversation).  In 2016, he started the EPSRC project MaDrIgAL, which investigates the notion of multidimensionality of communication in the context of statistical approaches to spoken dialogue systems.

Since December 2017, Simon is a lecturer in the AI Lab in Brussels.


Simon Keizer
Interaction Lab
School of Mathematical and Computer Sciences (MACS)Heriot-Watt University
Edinburgh, EH14 4AS
United Kingdom
Email: s<dot>keizer<at>hw<dot>ac<dot>uk

word cloud

Subpages (1): temp