Keynote Speaker 1: Tias Guns, Declarative Languages and Artificial Intelligence Lab., KU Leuven, Belgium

Title:
Constraint modeling and solving for data mining

Abstract:
The use of constraints is prevalent in data mining, most often to express background knowledge or feedback from the user. Constraints are also a well-studied formalism for modeling and solving combinatorial problems, as exemplified by the constraint programming community. In recent years, such constraint technology is increasingly used in the field of (symbolic) data mining. Many challenges exist though, at the level of modelling the problem (encodings, high-level and declarative languages, new primitives) as well as at the solving level (scalability, redundant constraints, search strategies). We review motivations and recent advances in the use of constraint programming for data mining problems.

Keynote Speaker 2: Guido Sanguinetti, School of Informatics, University of Edinburgh, UK

Title:
Machine learning methods for model checking in continuous time Markov chains

Abstract:
Model checking of temporal properties on stochastic processes is one of the major success stories of formal modelling. The applicability of model checking methods has been greatly extended by the availability of randomized, statistical algorithms such as Statistical Model Checking (SMC). Nevertheless, all of these methods require that a model is specified quantitatively, including a parametrization of the transition rates. Such prior knowledge is often unavailable in many important application fields such as systems biology. In this talk, I will discuss how an SMC procedure can be defined also for models with uncertain rates, by formalising the task in Bayesian framework and placing a non-parametric prior distribution over how the satisfaction probability of a formula depends on the model parameters. I will introduce the notion of Gaussian Process, and show how machine learning ideas can be used also for parameter synthesys and model design. If time permits, I'll also show how similar ideas can be used in more general reachability problems.

< Back to home