Kevin Sim

A lifelong learning hyper-heuristic method for bin packing

We describe a novel hyper-heuristic system that continuously learns over time to solve a combinatorial optimisation problem. The system continuously generates new heuristics and samples problems from its environment; and representative problems and heuristics are incorporated into a self-sustaining network of interacting entities inspired by methods in artificial immune systems. The network is plastic in both its structure and content, leading to the following properties: it exploits existing knowledge captured in the network to rapidly produce solutions; it can adapt to new problems with widely differing characteristics; and it is capable of generalising over the problem space. The system is tested on a large corpus of 3,968 new instances of 1D bin-packing problems as well as on 1,370 existing problems from the literature; it shows excellent performance in terms of the quality of solutions obtained across the datasets and in adapting to dynamically changing sets of problem instances compared to previous approaches. As the network self-adapts to sustain a minimal repertoire of both problems and heuristics that form a representative map of the problem space, the system is further shown to be computationally efficient and therefore scalable.

A hyper-heuristic classifier for one dimensional bin packing problems: improving classification accuracy by attribute evolution

A hyper-heuristic for the one dimensional bin packing problem is presented that uses an Evolutionary Algorithm (EA) to evolve a set of attributes that characterise a problem instance. The EA evolves divisions of variable quantity and dimension that represent ranges of a bins capacity and are used to train a k-nearest neighbour algorithm. Once trained the classifier selects a single deterministic heuristic to solve each one of a large set of unseen problem instances. The evolved classifier is shown to achieve results significantly better than are obtained by any of the constituent heuristics when used in isolation.

A hyper-heuristic ensemble method for static job-shop scheduling

We describe a new hyper-heuristic method NELLI-GP for solving job-shop scheduling problems (JSSP) that evolves an ensemble of heuristics. The ensemble adopts a divide-and-conquer approach in which each heuristic solves a unique subset of the instance set considered. NELLI-GP extends an existing ensemble method called NELLI by introducing a novel heuristic generator that evolves heuristics composed of linear sequences of dispatching rules: each rule is represented using a tree structure and is itself evolved. Following a training period, the ensemble is shown to outperform both existing dispatching rules and a standard genetic programming algorithm on a large set of new test instances. In addition, it obtains superior results on a set of 210 benchmark problems from the literature when compared to two state-of-the-art hyper-heuristic approaches. Further analysis of the relationship between heuristics in the evolved ensemble and the instances each solves provides new insights into features that might describe similar instances.

An improved immune inspired hyper-heuristic for combinatorial optimisation problems

The meta-dynamics of an immune-inspired optimisation system NELLI are considered. NELLI has previously shown to exhibit good performance when applied to a large set of optimisation problems by sustaining a network of novel heuristics. We address the mechanisms by which new heuristics are defined and subsequently generated. A new representation is defined, and a mutation-based operator inspired by clonal-selection introduced to control the balance between exploration and exploitation in the generation of new network elements. Experiments show significantly improved performance over the existing system in the bin-packing domain. New experiments in the job-scheduling domain further show the generality of the approach.

Generating single and multiple cooperative heuristics for the one dimensional bin packing problem using a single node genetic programming island model

Novel deterministic heuristics are generated using Single Node Genetic Programming for application to the One Dimensional Bin Packing Problem. First a single deterministic heuristic was evolved that minimised the total number of bins used when applied to a set of 685 training instances. Following this, a set of heuristics were evolved using a form of cooperative co-evolution that collectively minimise the number of bins used across the same set of problems. Results on an unseen test set comprising a further 685 problem instances show that the single evolved heuristic outperforms existing deterministic heuristics described in the literature. The collection of heuristics evolved by cooperative co-evolution outperforms any of the single heuristics, including the newly generated ones.

Learning to Solve Bin Packing Problems with an Immune Inspired Hyper-heuristic

Motivated by the natural immune system’s ability to defend the body by generating and maintaining a repertoire of antibodies that collectively cover the potential pathogen space, we describe an artificial system that discovers and maintains a repertoire of heuristics that collectively provide methods for solving problems within a problem space. Using bin-packing as an example domain, the system continuously generates novel heuristics represented using a tree-structure. An novel affinity measure provides stimulation between heuristics that cooperate by solving problems in different parts of the space. Using a test suite comprising of 1370 problem instances, we show that the system self-organises to a minimal repertoire of heuristics that provide equivalent performance on the test set to state-of-the art methods in hyper-heuristics. Moreover, the system is shown to be highly responsive and adaptive: it rapidly incorporates new heuristics both when entirely new sets of problem instances are introduced or when the problems presented change gradually over time.

A Combined Generative and Selective Hyper-heuristic for the Vehicle Routing Problem

Hyper-heuristic methods for solving vehicle routing problems (VRP) have proved promising on a range of data. The vast majority of approaches apply selective hyper-heuristic methods that iteratively choose appropriate heuristics from a fixed set of pre-defined low-level heuristics to either build or perturb a candidate solution. We propose a novel hyper-heuristic called GP-MHH that operates in two stages. The first stage uses a novel Genetic Programming (GP) approach to evolve high quality constructive heuristics; these can be used with any existing method that relies on a candidate solution(s) as its starting point. In the second stage, a perturbative hyper-heuristic is applied to candidate solutions created from the new heuristics. The new constructive heuristics are shown to outperform existing low-level heuristics. When combined with a naive perturbative hyper-heuristic they provide results which are both competitive with known optimal values and outperform a recent method that also designs new heuristics on some standard benchmarks. Finally, we provide results on a set of rich VRPs, showing the generality of the approach.

On Constructing Ensembles for Combinatorial Optimisation

Although the use of ensemble methods in machine-learning is ubiquitous due to their proven ability to outperform their constituent algorithms, ensembles of optimisation algorithms have received relatively little attention. Existing approaches lag behind machine-learning in both theory and practice, with no principled design guidelines available. In this article, we address fundamental questions regarding ensemble composition in optimisation using the domain of bin-packing as an example. In particular, we investigate the trade-off between accuracy and diversity, and whether diversity metrics can be used as a proxy for constructing an ensemble, proposing a number of novel metrics for comparing algorithm diversity. We find that randomly composed ensembles can outperform ensembles of high-performing algorithms under certain conditions and that judicious choice of diversity metric is required to construct good ensembles. The method and findings can be generalised to any metaheuristic ensemble, and lead to better understanding of how to undertake principled ensemble design.

On the life-long learning capabilities of a NELLI*: a hyper-heuristic optimisation system.

Real-world applications of optimisation techniques place more importance on finding approaches that result in acceptable quality solutions in a short time-frame and can provide robust solutions, capable of being modified in response to changes in the environment than seeking elusive global optima. We demonstrate that a hyper-heuristic approach NELLI* that takes inspiration from artifical immune systems is capable of life-long learning in an environment where problems are presented in a continuous stream and change over time. Experiments using 1370 bin-packing problems show excellent performance on unseen problems and that the system maintains memory, enabling it to exploit previously learnt heuristics to solve new problems with similar characteristics to ones solved in the past.

A real-world employee scheduling and routing application

We describe a hyper-heuristic application developed for a client to find quick, acceptable solutions to Workforce Scheduling and Routing problems. An interactive fitness function controlled by the user enables five different objectives to be weighted according to client preference. The application uses a real road network in order to calculate driving distances between locations, and is designed to integrate with a web-based application to access employee calendars.