China
Africa
Explore chapters and articles related to this topic
Literature Review
Published in Diego Oliva, Noe Ortega-Sánchez, Salvador Hinojosa, Marco Pérez-Cisneros, Modern Metaheuristics in Image Processing, 2022
Diego Oliva, Noe Ortega-Sánchez, Salvador Hinojosa, Marco Pérez-Cisneros
The topic of hyper-heuristics includes approaches that utilize a higher-level heuristic to select a low-level heuristic that can be applied for a specific problem. In the context of image thresholding, the hyper-heuristic-based methods will generate a particular heuristic designed for each image to be segmented. This kind of approach might fall into the last category, but we consider that there is great potential in this area to highlight it as a separate category. For example, we can observe the article proposed by Elaziz et al. (2019) where they present a Swarm Selection methodology that uses a Differential Evolution algorithm to select the best type of swarm-based algorithm that can be applied over a specific image. The study includes 10 popular algorithms working with Otsu as an objective function (Abd Elaziz et al., 2019). Also, the article by Elaziz et al. (2020) proposes a hyperheuristic framework based on the Genetic Algorithm (GA) designed to select the best combination of metaheuristics for a given set of images. Table 2.2 shows some examples of hyperheuristics used in multilevel thresholding.
Introduction
Published in Joseph Y.-T. Leung, Handbook of SCHEDULING, 2004
Burke and Petrovic (2002) published a paper which explored some research trends in automated timetabling. In that paper we briefly discussed hyper-heuristics as a promising timetabling research theme. A hyper-heuristic can be described as a “heuristic to choose a heuristic.” See Burke et al. (2003e) for a detailed discussion about hyper-heuristics. Ross et al. (1998) argued that a future way forward in timetabling research (in terms of genetic algorithms) might be to investigate genetic algorithms to choose the most appropriate heuristic for a problem rather than to operate directly on the problem. Indeed, Terashima-Marin et al. (1999) explored a genetic algorithm approach to the evolution of strategies in examination timetabling.
Feature selection approach for evolving reactive scheduling policies for dynamic job shop scheduling problem using gene expression programming
Published in International Journal of Production Research, 2023
Salama Shady, Toshiya Kaihara, Nobutada Fujii, Daisuke Kokuryo
Designing DRs is a tedious process that begins with formulating candidate rules and evaluating their performance on some problem instances. Accordingly, the structure of the rules is modified based on the insights gained. This cycle is repeated until the desired performance is achieved, which requires a significant amount of time, experience, and code effort (Branke et al. 2016). Therefore, hyper-heuristics have been proposed by incorporating machine learning techniques to automate the design of heuristics to solve computational search problems. Hyper-heuristics are defined as a high-level search methodology that explores the search space of low-level heuristics rather than the search space of solutions to the underlying problem (Burke et al. 2013). Regarding the automated generation of scheduling policies, Genetic Programming (GP) has been one of the most widely used approaches (Tay and Ho 2008). GP offers several advantages over other hyper-heuristics methods, such as flexible representation, a powerful search mechanism, and the availability of multi-objective optimization methods (Nguyen, Zhang, and Tan 2017). The tree-based GP representation is commonly used to evolve DRs in previous studies, e.g. (Shady et al. 2020b; Jakobović and Budin 2006; Shady et al. 2020a).
Simulated-annealing-based hyper-heuristic for flexible job-shop scheduling
Published in Engineering Optimization, 2022
Kelvin Ching Wei Lim, Li-Pei Wong, Jeng Feng Chin
The selection hyper-heuristic operates on an heuristic search space where a suitable heuristic is selected to solve a problem. Applying a fixed heuristic throughout could be disadvantageous, as the performance of heuristics is problem-dependent. Hence, instead of one, multiple heuristics could be selected (Garza-Santisteban et al.2019b). The idea is to exploit the strengths of each heuristic by applying the heuristic consecutively whenever a scheduling decision is needed (Kheiri and Keedwell 2015; Kheiri et al.2016). Such an approach is deemed to be less feasible on problem-dependent heuristics (e.g. dispatching rules). Another approach proposed by Garza-Santisteban et al. (2019b) includes the use of problem state features to facilitate the application of multiple heuristics, namely an heuristic scheme (HS). In this approach, the choice of heuristic is determined by a mathematical model based on a comparison between the current problem state and the pre-defined problem state in the HS. The choice of heuristic that corresponds to a particular problem state makes it more efficient in solving a problem. While Garza-Santisteban et al. (2019b) have experimented with the HS formulation on the JSP, the present research applies the design of the HS with adaptations to the formulation of problem state features so that it becomes applicable to the FJSP.
A novel feature selection for evolving compact dispatching rules using genetic programming for dynamic job shop scheduling
Published in International Journal of Production Research, 2022
Salama Shady, Toshiya Kaihara, Nobutada Fujii, Daisuke Kokuryo
The manual design of dispatching rules starts by defining a set of job shop attributes, then trying to obtain the proper combination of these terminals in a mathematical form. Later, all the candidate rules are hard-coded and evaluated using a Discrete Event Simulation (DES) model. This cycle is repeated several times to obtain the best rule that can optimise a performance measure under certain job shop settings (Branke et al. 2016). Recently, a hyper-heuristic approach has been proposed to automate the process of selecting or generating heuristics from a set of simple heuristics or components of such heuristics. In other words, hyper-heuristics aims at raising the level of generality of search methods by discovering the right heuristic (method) in each situation instead of solving the problem directly. Regarding the automatic design of scheduling rules, GP has been shown to be a promising hyper-heuristic approach that dominates other machine learning methods (Branke, Hildebrandt, and Scholz-Reiter 2015).