Ant Colony Hyper-heuristics for Graph Colouring Nam Pham ASAP Group, Computer Science School University of Nottingham Overview Hyper-heuristic Framework Problem Description Hyper-heuristic design for the problem An ant colony hyper-heuristic approach and experimental results Future works Nam Pham Hyper-heuristic “Heuristics that choose heuristics” High level heuristics: Meta-heuristics Choice Function Ant Algorithm Case-based Reasoning … Low level heuristics: different moving strategies, constructive heuristics … Nam Pham Hyper-heuristic Framework Nam Pham Graph Colouring Problem Assignment of “colours” to vertices in a graph Adjacent vertices have different colours Objective: minimise the number of required colours Nam Pham Hyper-heuristic Design Constructive hyper-heuristics Search for sequence of heuristics [Ross 2002] Each heuristic is applied for colouring one vertex Evaluation function is defined as the number of required colours when applying heuristic sequence Nam Pham Graph Example 1 Heuristic 1 (H1) Heuristic 2 (H2) Heuristic 3 (H3) 2 8 3 7 4 6 5 Nam Pham Search Space of Heuristic Sequences We are looking for a heuristic sequence that produces smallest number of used colours Decisions H1 H2 H3 Sequence 1 2 3 Nam Pham 4 5 6 7 8 Ant Colony Hyper-heuristics Ant algorithms are well-known if used as low level heuristics There are only two papers using ant algorithms as hyper-heuristics so far (reference at the end) Nam Pham Ant Colony Hyper-heuristics Ant algorithm is well-known if used as a low level heuristic There are only two papers using ant algorithm as hyper-heuristic so far (reference at the end) Nam Pham Ant Colony Hyper-heuristics Ant algorithm is well-known if used as a low level heuristic There are only two papers using ant algorithm as hyper-heuristic so far (reference at the end) Nam Pham Ant Colony Hyper-heuristics Ant algorithm is well-known if used as a low level heuristic There are only two papers using ant algorithm as hyper-heuristic so far (reference at the end) Nam Pham Experiment Heuristics employed include: Largest Degree First (LD) Largest Colour Degree First (LCD) Least Saturation Degree First (SD) University of Toronto Benchmark Data ftp://ftp.mie.utoronto.ca/pub/carter/testprob Nam Pham Results LD LCD SD Ant Algorithm HH Best known Car91 37 35 31 29 28 Car92 32 33 33 29 28 Ear83 26 25 23 22 22 Hec92 19 20 19 17 17 Kfu93 20 21 20 19 19 Lse91 17 18 17 17 17 Pur93 36 40 35 32 35 Rye93 22 23 22 21 21 Sta83 13 13 13 13 13 Tre92 23 26 23 20 20 Uta92 32 33 32 30 30 Ute92 10 10 10 10 10 Yor83 22 24 22 19 19 Nam Pham Future works Compare ant colony hyper-heuristic with other population based hyper-heuristics – evolutionary algorithms, genetic algorithm, swarm intelligence… Do research on characteristics of heuristic search space Expand to exam timetabling problem Nam Pham Reference Burke, E.K., Kendall, G., Landa Silva, J.D., O'Brien, R.F.J., Soubeiga, E.: An ant algorithm hyperheuristic for the project presentation scheduling problem. Cuesta-Cañada, A., Garrido, L., Terashima-Marín, H.: Building Hyper-heuristics Through Ant Colony Optimization for the 2D Bin Packing Problem. Nam Pham