- 152 Downloads
It is great to start 2018 with this issue of 8 papers. To cater to the increasing number of submissions, each print issue of this journal will now have an increased page budget to accommodate more papers. It is expected that the number of submissions will continue to increase, which inadvertently would result in greater effort in managing the review process. To maintain the quality of the papers selected for publication, manuscripts received before even going through a proper review process must demonstrate an acceptable level of language proficiency. Technical relevance in terms of scope and objectives would also be used as a first-cut evaluation of suitability.
We start this issue with a paper on sentiments classification. Collecting and processing information for multimedia websites involves dealing with huge amount of information changing at a very high speed. To achieve high processing and computational throughput, Tran & Cambria leveraged on the processing speed of extreme learning machine (ELM) and graphics processing unit (GPU) to overcome the limitations of standard learning algorithms and central processing unit (CPU). They applied this in real-time multimodal sentiment analysis, i.e., harvesting sentiments from Web videos by taking into account audio, visual and textual modalities. To facilitate sentiment classification, they described the notion of sentic memes; basic units of sentiment whose combination can potentially describe the full range of emotional experiences. Based on sentiment annotated dataset generated from video reviews, they achieved good accuracy. The processing speed was reported to be several orders of magnitude improvement in features extraction compared to CPU-based counterparts.
The second paper deals with the issue of mining fuzzy rules. The advantage of fuzzy rules is that the representation of domain knowledge is “transparent” and the reasoning process is traceable. However, the issue of finding or formulating the rules and membership functions can be tedious and non-trivial, at times ad hoc. Ting et. al. present a memetic approach for deriving suitable and relevant membership functions in fuzzy associative rule mining. To achieve the desired objectives, their memetic computation method uses a novel chromosomal representation that incorporates information of the membership functions structure and was able to achieve good performance in terms of mining fuzzy rules.
In problem-solving, any a priori knowledge about the domain can be exploited to improve the performance of basic algorithms or metaheuristics. Genetic algorithms on its own obviously have its limits. In their paper, Rezoug et. al. applied GA to solve the multidimensional knapsack problem. They applied problem-data pre-analysis using an efficiency based method to extract useful information about the problem. Based on the a priori information, they used it in their two-stage guided GA to drive the generation of initial population as well as fitness evaluation function. The performance of their approach was evaluated on a set of well-known benchmark data with competitive results.
The next paper on extreme learning machine (ELM) is the work by Zhang et. al. that describes the kernel online sequential ELM, an extension of the least-squares ELM framework. The objective is to deal with time varying data. They incorporated a mechanism to “forget” old observations by means of a dynamic sliding window to limit the amount of active training data. From their empirical results, they reported that the approach showed improvement compared to other ELM-based approaches.
Yang et. al. describe their work on self-adjustment of parameters for visual background editor (ViBe) to suit the environment condition. The parameters control the number of samples chosen from the background template. They proposed two models to do auto-tuning of parameters; blink energy model and object probability model. Based on experimental results on benchmark datasets, they validated their approach in terms of dealing with dynamic backgrounds and preventing image corrosion for moving objects.
The paper by Guendouzi and Boukra is on feature selection. They proposed a new selection method, combining local neighbourhood search (LNS) and ensemblist discrete differential evolution (EDDE). There are two phases in their approach. The LNS improves features subset through a “destroy and repair” process. They defined an accuracy rate difference metric to sieve out irrelevant or redundant features for removal during the destruction process. In the second phase, the individuals resulting from LNS are used as inputs for the EDDE which attempts to find the best features subset in a single dimension space through the application of ensemblist operators on a set of features. They applied their method on machine learning datasets with competitive results in comparison to other established feature selection methods.
Combination of different strategies within a single problem-solving framework has often served as a backbone for innovative approaches in dealing with complex problems. In the next paper, a quantum-inspired immune clonal algorithm in solving combinatorial optimization problem is proposed. The algorithm of Shang el. al. combines artificial immune system with quantum computing qubit and superposition applied to solve large-scale capacitated arc routing problem. They showed that the approach produced results with competitive performance in terms of convergence rate and solution quality.
To wrap up this issue, the paper by Li et. al. presents an innovative decomposition-based chemical reaction optimization for multi-objective vehicle routing problem (VRP). The VRP is a computationally challenging combinatorial optimization problem with many different variants to cater to real world scenarios and constraints. A practical variant of the vehicle routing problem (VRP), called VRP, with simultaneous delivery and pickup and time windows (VRPSDPTW) is a challenging combinatorial optimization problem that has five optimization objectives in transportation and distribution logistics. In their approach, a sub-problem is a chemical molecule, with molecules within the search domain divided into groups. Each molecule associated to several neighboring molecules. They designed operators of on-wall ineffective collision and inter-molecular ineffective collision for a local search, as well as operators of decomposition and synthesis to enhance global convergence. Comparing with two different algorithms based-on hypervolume performance measures they show results that outperform the other algorithms in most benchmarks.
The 8 papers compiled for this issue make for a good start with promising development in the field of search and optimization. I am looking forward to an exciting and productive year ahead, hopeful of more submissions that challenge the norms; pushing established boundaries or markers in computational problem-solving. I gratefully acknowledge the hard work and dedication of the reviewers in evaluating the manuscripts submitted for publication and look forward to their continued support in this journal. Also to the Editors assigned to manage the review of the papers, my sincere thanks for the job well done.