Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Graham Kendall Contents The University of Nottingham • What is a hyper-heuristic? gxk@cs.nott.ac.uk • What motivates hyper-heuristic http://www.cs.nott.ac.uk/~gxk research? research? • Hyper-heuristic Methods • Observations and Future Potential • Questions/Discussions Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies What is a hyper-heuristic? What is the difference between a hyper- heuristic and a meta-heuristic? • Simple Idea: Heuristics to choose • All the term hyper-heuristic says is: “ Operate on heuristics a search space of heuristics ”. • Operates on a search space of heuristics • We have a high level search method (the hyper- W h hi h l l h th d (th h rather than directly on a search space of heuristic) - which may (or may not) be a meta- solutions heuristic • Meta-heuristics have been used with some success as hyper-heuristics, as have other approaches such as case based reasoning 1
Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies What is the difference between a Different Search Spaces hyper-heuristic and a meta- heuristic? Hyper-heuristic Operates upon • A hyper-heuristic searches through a space of heuristics (which potentially could be of heuristics (which, potentially, could be Heuristics meta-heuristics) • Most meta-heuristic implementations Operates upon operate directly on a search space of potential solutions Potential Solutions Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Contents Motivations behind hyper-heuristic research • What is a hyper-heuristic? • What motivates hyper-heuristic • What is our strategic research vision? research? • What game to play? Wh t t l ? • Hyper-heuristic Methods • Observations and Future Potential • Questions/Discussions 2
Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies The “Best Quality Solution” Game The “Best Quality Solution” Game • Apply methodologies to benchmarks • We have a problem (e.g. exam timetabling) • Compare with other “players” • The goal is to get better quality • The goal is to “get better quality • We have a set of benchmark problems W h t f b h k bl solutions” than the other players • We develop new methodologies (ever • Good Game more sophisticated) Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Consequences of “Best Quality Solution” Game The “Best Quality Solution” Game • Handcrafted bespoke decision support methodologies • Made to Measure – NOT off the peg p g • Rolls-Royce systems • We compare Rolls-Royces with Bentleys Benchmark Instances with Mercedes with Ferraris…….. And it works e.g. Exam Timetabling 3
Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies A new game to play: Many Walls? A New Game? • Raising the level of generality • What about Ford Model Ts’? • Still want to get as high up the wall as • Do we have the technology to mass produce possible ……… BUT……… p decision support systems? decision support systems? • We want to be able to operate on as • Develop decision support systems that are many different walls as possible off the peg? • The goal is to increase the number of • Can we develop the ability to automatically walls you can operate on – while still work well on different problems? getting acceptably high up each wall Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Raising the Level of Generality Evaluating Different Methods • It IS a new game • Cannot sensibly compare a Model T to a Rolls Royce – different ff functions for different clientele • Raising the level of generality is the goal One Method 4
Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Evaluating Different Methods A Grand Challenge • Still want solution quality to be as high as • Automating the heuristic design process possible though • Deeper understanding of how high we • Good Enough – Soon Enough – Cheap Good Enough Soon Enough Cheap can raise the level of generality: What can raise the level of generality: What Enough are the limits? Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Hyper-heuristics A Grand Challenge • Motivated by raising the level of generality The General Solver Doesn’t exist…. • Role to play in BOTH games Significant scope for future research g p More General These situations exist Problem Specific Solvers 5
Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Contents Intellectual Roots • What is a hyper-heuristic? • Hyper-heuristic research has many of its • What motivates hyper-heuristic roots in the mid-nineties research? research? • But can be traced back to the 60’s • Hyper-heuristic Methods • Observations and Future Potential • Questions/Discussions Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Intellectual Roots Intellectual Roots • Fisher H. and Thompson G.L. Probabilistic Learning • H-L Fang, P.M.Ross and D.Corne. A Promising Hybrid Combinations of Local Job-shop Scheduling Rules. In Factory Scheduling Conference, Carnegie Institute of Technology, May GA/Heuristic Approach for Open-Shop Scheduling 10-12 , 1961 Problems'', in Proceedings of ECAI 94: 11 th European • Fisher H. and Thompson G.L. Probabilistic Learning Conference on Artificial Intelligence , A. Cohn (ed), pp 590- Combinations of Local Job-shop Scheduling Rules. Ch 15,:225- 251, Prentice Hall, New Jersey , 1963 251 Prentice Hall New Jersey 1963 594 John Wiley and Sons Ltd 1994 594, John Wiley and Sons Ltd, 1994 • Crowston W.B., Glover F., Thompson G.L. and Trawick J.D. Probabilistic and Parameter Learning Combinations of Local Job The chromosome was a set of heuristics that was chosen Shop Scheduling Rules. ONR Research Memorandum, GSIA, Carnegie Mellon University, Pittsburgh , (117), 1963 to schedule a job on a machine. As opposed to the “normal” method of having the chromosome as The learning mechanism used probabilistic weightings representing a set of jobs to be scheduled in a given order. of low level heuristics which represented scheduling rules 6
Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies A Hyper-heuristic Framework Intellectual Roots Hyper-heuristic • Hart E, Ross P. and Nelson J.A.D. Solving a Real World Problem using an Evolving Heuristically Driven Schedule Data flow Builder. Evolutionary Computing 6(1):61-80 , 1998 • Hart E, Ross P. and Nelson J.A.D. Scheduling Chicken Domain Barrier Catching: An Investigation into the Success of a Genetic C t hi A I ti ti i t th S f G ti Data flow Algorithm on a Real World Scheduling Problem. Annals of Operations Research 92:363-380 , 1999 Set of low level heuristics H 1 H 2 H n Problem is decomposed into two sub-problems, each being …… solved with a GA. The chromosomes represent heuristics. Evaluation Function Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Choice function Hyper-heuristic Choice Function Sales Summit Scheduling Hyper-heuristic f1 + f2 + f3 Low level heuristics are based on three types of Data flow How well has each heuristic done + neighbourhood moves : Domain Barrier How well have pairs of heuristics done + How well have pairs of heuristics done + • Add / Remove one delegate to / from the current solution Time since last called Data flow • Add / remove a meeting to / from the current solution Set of low level (random, 1 st improving, best improving, etc) Applied to sales summit scheduling, heuristics CS&IT presentations, nurse rostering • Remove excess of meetings from an overloaded delegate …… H 2 H n H 1 Evaluation Function 7
Recommend
More recommend