There are no integers in discrete optimisation

Speaker: Professor Peter Stuckey
OPTIMA _ Monash University

Title: There are no integers in discrete optimisation

Abstract:
Discrete optimisation problems make decisions from a finite set of choices. They encompass many important problem classes such as scheduling, rostering and resource allocation. MiniZinc is a leading modelling language for discrete optimisation. It allows the expression of discrete optimisation problems succinctly using high level global constraints, and automatically translates them to run on constraint programming (CP), mixed integer programming (MIP), Boolean satisfiability (SAT), SAT modulo theories (SMT), and local search solvers. Integers are a key type in MiniZinc since they are used to represent all the finite decisions made during solving.  But in the latest development version of MiniZinc, we recommend never using integers in models. Why? Finding errors in discrete optimisation models can be very challenging. In the worst case when a solver simply returns no answer, we don’t know if this is because the problem we want to ask is too hard (for this solver) or the problem we actually asked (because of errors in the model) is too hard. Looking through solver traces of millions of events to find a problem is very hard work, and indeed there may be no error.  So strong typing is important for modelling languages, the more the compiler discovers modelling errors, the less challenging debugging is required. In this talk I will talk about how we ensure that no (pure) integers appear in MiniZinc models by using enumerated types and type extensions, as well as unit types, to clearly differentiate between the different kinds of integers appearing in a model.

Bio:
Peter Stuckey from Monash University has a world-leading research program in constraint programming. He led the G12 project, one of the largest projects at NICTA, peaking at 25 researchers, culminating in the Opturion spinout company, delivering optimisation solutions to commercial customers. He has been involved in 3 ARC Linkage Projects and other industry contract research projects, working in Energy, Security, Resources and Transport. His research expertise in optimisation broadly covers modelling languages and model transformation, solving using Artificial Intelligence (AI) and Operations Research (OR) technology, and using machine learning methods in concert with optimisation. For the global influence and uptake of his research, he was awarded the 2010 Google Australia Eureka Prize for Innovation in Computer Science and honoured in 2019 with a Fellowship of the prestigious Association for AI Advancement.

More information: research.monash.edu/en/persons/peter-stuckey