Shahriar Asta

A tensor-based selection hyper-heuristic for cross-domain heuristic search

Presented a novel method to improve the performance of a hyper-heuristic approach.Represented the trail of a hyper-heuristic mixing heuristics as a 3rd-order tensor.Analyzed tensors to discover the links between heuristics and the hyper-heuristic.Obtained competitive results to the best known hyper-heuristics across 6 domains. Hyper-heuristics have emerged as automated high level search methodologies that manage a set of low level heuristics for solving computationally hard problems. A generic selection hyper-heuristic combines heuristic selection and move acceptance methods under an iterative single point-based search framework. At each step, the solution in hand is modified after applying a selected heuristic and a decision is made whether the new solution is accepted or not. In this study, we represent the trail of a hyper-heuristic as a third order tensor. Factorization of such a tensor reveals the latent relationships between the low level heuristics and the hyper-heuristic itself. The proposed learning approach partitions the set of low level heuristics into two subsets where heuristics in each subset are associated with a separate move acceptance method. Then a multi-stage hyper-heuristic is formed and while solving a given problem instance, heuristics are allowed to operate only in conjunction with the associated acceptance method at each stage. To the best of our knowledge, this is the first time tensor analysis of the space of heuristics is used as a data science approach to improve the performance of a hyper-heuristic in the prescribed manner. The empirical results across six different problem domains from a benchmark indeed indicate the success of the proposed approach.

An apprenticeship learning hyper-heuristic for vehicle routing in HyFlex

Apprenticeship learning occurs via observations while an expert is in action. A hyper-heuristic is a search method or a learning mechanism that controls a set of low level heuristics or combines different heuristic components to generate heuristics for solving a given computationally hard problem. In this study, we investigate into a novel apprenticeship-learning-based approach which is used to automatically generate a hyper-heuristic for vehicle routing. This approach itself can be considered as a hyper-heuristic which operates in a train and test fashion. A state-of-the-art hyper-heuristic is chosen as an expert which is the winner of a previous hyper-heuristic competition. Trained on small vehicle routing instances, the learning approach yields various classifiers, each capturing different actions that the expert hyper-heuristic performs during the search process. Those classifiers are then used to produce a hyper-heuristic which is potentially capable of generalizing the actions of the expert hyper-heuristic while solving the unseen instances. The experimental results on vehicle routing using the Hyper-heuristic Flexible (HyFlex) framework shows that the apprenticeship-learning-based hyper-heuristic delivers an outstanding performance when compared to the expert and some other previously proposed hyper-heuristics.

Batched Mode Hyper-heuristics

A primary role for hyper-heuristics is to control search processes based on moves generated by neighbourhood operators. Studies have shown that such hyper-heuristics can be effectively used, without modification, for solving unseen problem instances not only from a particular domain, but also on different problem domains. They hence provide a general-purpose software component to help reduce the implementation time needed for effective search methods. However, hyper-heuristic studies have generally used time-contract algorithms i.e. a fixed execution time and also solved each problem instance independently. We consider the potential gains and challenges of a hyper-heuristic being able to treat a set of instances as a batch; to be completed within an overall joint execution time. In batched mode, the hyper-heuristic can freely divide the computational effort between the individual instances, and also exploit what it learns on one instance to help solve other instances.

A tensor based hyper-heuristic for nurse rostering

Nurse rostering is a well-known highly constrained scheduling problem requiring assignment of shifts to nurses satisfying a variety of constraints. Exact algorithms may fail to produce high quality solutions, hence (meta)heuristics are commonly preferred as solution methods which are often designed and tuned for specific (group of) problem instances. Hyper-heuristics have emerged as general search methodologies that mix and manage a predefined set of low level heuristics while solving computationally hard problems. In this study, we describe an online learning hyper-heuristic employing a data science technique which is capable of self-improvement via tensor analysis for nurse rostering. The proposed approach is evaluated on a well-known nurse rostering benchmark consisting of a diverse collection of instances obtained from different hospitals across the world. The empirical results indicate the success of the tensor-based hyper-heuristic, improving upon the best-known solutions for four of the instances.

CHAMP: Creating Heuristics via Many Parameters for online bin packing

The online bin packing problem is a well-known bin packing variant which requires immediate decisions to be made for the placement of a lengthy sequence of arriving items of various sizes one at a time into fixed capacity bins without any overflow. The overall goal is maximising the average bin fullness. We investigate a ‘policy matrix’ representation which assigns a score for each decision option independently and the option with the highest value is chosen for one dimensional online bin packing. A policy matrix might also be considered as a heuristic with many parameters, where each parameter value is a score. We hence investigate a framework which can be used for creating heuristics via many parameters. The proposed framework combines a Genetic Algorithm optimiser, which searches the space of heuristics in policy matrix form, and an online bin packing simulator, which acts as the evaluation function. The empirical results indicate the success of the proposed approach, providing the best solutions for almost all item sequence generators used during the experiments. We also present a novel fitness landscape analysis on the search space of policies. This study hence gives evidence of the potential for automated discovery by intelligent systems of powerful heuristics for online problems; reducing the need for expensive use of human expertise.

A Tensor Analysis Improved Genetic Algorithm for Online Bin Packing

Mutation in a Genetic Algorithm is the key variation operator adjusting the genetic diversity in a population throughout the evolutionary process. Often, a fixed mutation probability is used to perturb the value of a gene. In this study, we describe a novel data science approach to adaptively generate the mutation probability for each locus. The trail of high quality candidate solutions obtained during the search process is represented as a 3rd order tensor. Factorizing that tensor captures the common pattern between those solutions, identifying the degree of mutation which is likely to yield improvement at each locus. An online bin packing problem is used as an initial case study to investigate the proposed approach for generating locus dependent mutation probabilities. The empirical results show that the tensor approach improves the performance of a standard Genetic Algorithm on almost all classes of instances, significantly.

Memetic algorithms for Cross-domain Heuristic Search

Hyper-heuristic Flexible Framework (HyFlex) is an interface designed to enable the development, testing and comparison of iterative general-purpose heuristic search algorithms, particularly selection hyper-heuristics. A selection hyper-heuristic is a high level methodology that coordinates the interaction of a fixed set of low level heuristics (operators) during the search process. The Java implementation of HyFlex along with different problem domains was recently used in a competition, referred to as Cross-domain Heuristic Search Challenge (CHeSC2011). CHeSC2011 sought for the best selection hyper-heuristic with the best median performance over a set of instances from six different problem domains. Each problem domain implementation contained four different types of operators, namely mutation, ruin-recreate, hill climbing and crossover. CHeSC2011 including the competing hyper-heuristic methods currently serves as a benchmark for hyper-heuristic research. Considering the type of the operators implemented under the HyFlex framework, CHeSC2011 could also be used as a benchmark to empirically compare the performance of appropriate variants of the evolutionary computation methods across a variety of problem domains for discrete optimisation. In this study, we investigate the performance and generality level of generic steady-state and transgenerational memetic algorithms which hybridize genetic algorithms with hill climbing across six problem domains of the CHeSC2011 benchmark.

A self-adaptive Multimeme Memetic Algorithm co-evolving utility scores to control genetic operators and their parameter settings

Graphical abstractDisplay Omitted HighlightsA self-adaptive MA is introduced using a different encoding scheme to a generic MMA.A utility score for each option is maintained within each individual in the population.Utility scores are updated using a strategy inspired by Reinforcement Learning.The new MA is shown to outperform a generic MA on average over six benchmark problems.Good results are also observed when compared to existing methods from the literature. Memetic algorithms are a class of well-studied metaheuristics which combine evolutionary algorithms and local search techniques. A meme represents contagious piece of information in an adaptive information sharing system. The canonical memetic algorithm uses a fixed meme, denoting a hill climbing operator, to improve each solution in a population during the evolutionary search process. Given global parameters and multiple parameterised operators, adaptation often becomes a crucial constituent in the design of MAs. In this study, a self-adaptive self-configuring Steady-state Multimeme Memetic Algorithm (SSMMA) variant is proposed. Along with the individuals (solutions), SSMMA co-evolves memes, encoding the utility score for each algorithmic component choice and relevant parameter setting option. An individual uses tournament selection to decide which operator and parameter setting to employ at a given step. The performance of the proposed algorithm is evaluated on six combinatorial optimisation problems from a cross-domain heuristic search benchmark. The results indicate the success of SSMMA when compared to the static MAs as well as widely used self-adaptive Multimeme Memetic Algorithm from the scientific literature.

Evolutionary Cross-Domain Hyper-Heuristics

Designing effective algorithms to solve computational problems is difficult and time-consuming. The standard methodology for designing such algorithms is “top-down”. This process breaks down large problems into more understood components and eventually identifies problem-specific operators that algorithms need to use to solve the given problem. Often, restrictive assumptions have to be made about the use of operators within an algorithm.We argue that it is desirable to automate this process. A wider of range of possible algorithms can be generated automatically and new TSP solvers could be discovered, in a reasonable amount of time and without the restrictions imposed by the human mind. We focus on evolving a fixed sequence of operators inside the loop of a Memetic Algorithm, using an innovative automatic algorithm creation method. We are proposing to extract and hard-code these evolved algorithms in new independent solvers, to find good solutions to a chosen problem.

Soft morphological filter optimization using a genetic algorithm for noise elimination

Digital image quality is of importance in almost all image processing applications. Many different approaches have been proposed for restoring the image quality depending on the nature of the degradation. One of the most common problems that cause such degradation is impulse noise. In general, well known median filters are preferred for eliminating different types of noise. Soft morphological filters are recently introduced and have been in use for many purposes. In this study, we present a Genetic Algorithm (GA) which combines different objectives as a weighted sum under a single evaluation function and generates a soft morphological filter to deal with impulse noise, after a training process with small images. The automatically generated filter performs better than the median filter and achieves comparable results to the best known filters from the literature over a set of benchmark instances that are larger than the training instances. Moreover, although the training process involves only impulse noise added images, the same evolved filter performs better than the median filter for eliminating Gaussian noise as well.