GECCO 2013 Workshop Summary

Workshop on Problem Understanding and Real-World Optimisation

The workshop was jointly organised by the ROLL team (Emma Hart and Kevin Sim) and Dr Kent McClymont of the University of Exeter. The morning was devoted to looking at aspects of real-word optimisation – why real-world problems are hard, efforts to facilitate solving problems and shared experiences from people who have worked at closely in industry.

Real World Problems

Unfortunately, the invited speaker, Tim Pigden of Optrak Logistics was stricken with a bout of gout and sadly couldn’t make it to the workshop. However, technology prevailed and Tim was able to deliver his talk from his bed over Skype (the empty stage generating a few odd looks from people passing the room!). Based on many years of experience of supply software to the vehicle routing industry, Tim outlined some of the many real constraints associated with vehicle routing that require practitioners to understand demand, travel, real-time and people constraints. To summarise Tim’s words – academics creating benchmarks need to understand that:

        • people place orders
        • tractors, trailers and drivers need to be treated properly
        • linear capacity constraints are best suited to liquids

This was illustrated with many practical examples, citing examples of deliveries of newspapers, petrol and lubricant, pipes and beer! A great many constraints were highlighted that it seems safe to say most of the audience had never considered in their own research on VRP and provided much food for thought. One of the things we intend to pursue as part of the ROLL project is the creation of a more realistic set of VRP benchmarks, that capture at least some of the many constraints that Tim raised.

Tim’s slides can be downloaded here as a powerpoint file.

This talk was followed by a presentation from Neil Urquhart, who described recent work in using GIS such as Google Maps or Open Street Map with VRPs, accounting for real road-network layouts rather than using proxy measures such as Euclidean distance. With the increasing availability of such tools, combining proper mapping with real problem descriptions as outlined by Tim surely has the potential to close some of the gap between academic VRP research and the real-world.

Tim Pigden’s talk was nicely complemented by a talk from Bogdan Filipic from the Franz Josef Institute, Slovenia, who shared experiences of working within industry in two specific cases; tuning the process parameters for casting of steel, and scheduling interruptions in a car factory to minimise energy costs. Bogdan concluded with a number of tips for those embarking on a joint work with industrial partners that will be of benefit to many people.

Also relating to real-world applications, Dr Gabriela Ochoa discussed some exciting opportunities for applying optimisation research to the software-engineering domain, outlining new research in finding and fixing bugs in software. She also described a recent research initiative Dynamic Adaptive Automated Software Engineering (DAASE), whose goal is to embed optimisation into deployed software to create self-optimising adaptive systems – definitely a real-word application for optimisation researchers to get their teeth into!

Finally, Ioannis Giagkiozis from the University of Sheffield presented Liger – an open source integrated optimization environment which is designed to be extensible and have a smooth learning curve so that it can be used by the non-expert in industry, thus addressing a current problem that many software packages designed to solve real-world problems are commercial or closed-source. The framework provides a number of
fully configurable algorithms, the facility to create new algorithms and a visual exploration tool – essential for decision makers. Making tools such as Liger available will hopefully help address one of the current problems for industry, i.e. that adopting novel technologies often requires expensive and hard-to-find access to an academic expert.

Problem Understanding
The morning’s workshop focusing on aspects of real-world problems was nicely complemented by the afternoon’s discussions, which examined current methods of understanding problem landscapes. Although this research uses benchmark problems to gain some theoretical understanding into what makes certain landscapes difficult for optimisation algorithms, this kind of research is crucial in developing our understanding of when and where algorithms work. By combining this knowledge with an improved understanding of the nature of real-world problems, it should indeed to possible to make inroads into ‘bridging the gap’ between real-world and academic problems.