The studies of evolutionary algorithms can be classified according to the type of algorithm or the type of optimisation problem (OP). From the perspective of OPs, the domination landscape (DL) offers a way to characterise the difficulty or complexity of an OP. This paper proposes a new method called accompanying optimisation (AO), which focuses on making optimisation easier by leveraging properties of the OPs rather than primarily focusing on algorithm tuning. AO utilises an accompanying OP (aOP) that is DL-equivalent to the original OP, with both sharing the same DL. Based on the properties of aOPs, judgment theorems are introduced to determine when an aOP is globally or locally DL-equivalent to the OP. The objective of this study is to reduce the difficulty level in solving complex optimisation problems. During the optimisation of an aOP, local knowledge is gained. This knowledge is then transferred to enhance the performance of algorithms.

As an emerging network model, spiking neural networks (SNNs) have aroused significant research attentions in recent years. However, the energy-efficient binary spikes do not augur well with gradient descent-based training approaches. Surrogate gradient (SG) strategy is investigated and applied to circumvent this issue and train SNNs from scratch. Due to the lack of well-recognized SG selection rule, most SGs are chosen intuitively. We propose the parametric surrogate gradient (PSG) method to iteratively update SG and eventually determine an optimal surrogate gradient parameter, which calibrates the shape of candidate SGs. In SNNs, neural potential distribution tends to deviate unpredictably due to quantization error. We evaluate such potential shift and propose methodology for potential distribution adjustment (PDA) to minimize the loss of undesired pre-activations. Experimental results demonstrate that the proposed methods can be readily integrated with backpropagation through time (BPTT) algorithm and help modulated SNNs to achieve state-of-the-art performance on both static and dynamic dataset with fewer timesteps.

High-order problems pose significant challenges for evolutionary algorithms (EAs) to optimize. To mitigate this, a deep hybrid transfer learning EA (DHTL-EA) is proposed. DHTL-EA works by transferring both the model and the optima from a corresponding low-order problem. Here, a deep neural network is adopted to model both the low-order and high-order problems, and the training data are derived from historical evolutionary data. The theoretical basis of DHTL-EA is well-supported by the transferability of solutions among domination-landscape-equivalent problems. The transfer to high-order problem is achieved by retraining the last fully connected layer of the deep neural network model for the target problem. Experiments on two groups of problems validated that DHTL-EA is effective on high-order problems. Further testing on other benchmark functions demonstrated its competitive performance.

Artificial Neural Network (ANN) has served as an important pillar of machine learning which played a crucial role in fueling the robust artificial intelligence (AI) revival experienced in the last few years. Inspired by the biological brain architecture of living things, ANN has shown widespread success in pattern recognition, data analysis and classification tasks. Among the many models of neural networks conceptualized and developed over the years, the Spiking Neural Network (SNN) which was initiated in 1996 has shown great promise in the current push towards compact embedded AI. By combining both spatial and temporal information as features in the training and testing process, many inherent shortcomings of traditional ANNs can be overcome. With temporal features and event-driven updating of the network, SNNs hold the potential of improving computational and energy efficiency. In SNN, the most basic signal carrier element is the spike, bringing about a revolution in neural network weights updating compared to traditional methods that are widely applied in ANNs. In literature, there have been numerous SNN weights updating algorithms developed in recent years. With the active and dynamic research work on SNN, a consolidation of the state-of-the-art SNN research is beneficial. This paper is aimed at reviewing and surveying the current status of research pertaining to SNN, in particular highlighting the various novel SNN training techniques coupled with an objective comparison of the techniques. SNN applications and associated neuromorphic hardware systems are also covered in this survey, with some thoughts on the challenges in developing new SNN training algorithms and discussion on potential future research trends are presented in this survey.

Spiking Neural Networks (SNNs) have shown substantial promise in processing spatio-temporal data, mimicking biological neuronal mechanisms, and saving computational power. However, most SNNs use fixed model regardless of their locations in the network. This limits SNNs’ capability of transmitting precise information in the network, which becomes worse for deeper SNNs. Some researchers try to use specified parametric models in different network layers or regions, but most still use preset or suboptimal parameters. Inspired by the neuroscience observation that different neuronal mechanisms exist in disparate brain regions, we propose a new spiking neuronal mechanism, named learnable thresholding, to address this issue. Utilizing learnable threshold values, learnable thresholding enables flexible neuronal mechanisms across layers, proper information flow within the network, and fast network convergence. In addition, we propose a moderate dropout method to serve as an enhancement technique to minimize inconsistencies between independent dropout runs. Finally, we evaluate the robustness of the proposed learnable thresholding and moderate dropout for image classification with different initial thresholds for various types of datasets. Our proposed methods produce superior results compared to other approaches for almost all datasets with fewer timesteps. Our codes are available at https://github.com/sq117/LTMD.git.

Indoor positioning is a key technology enabler for various smart systems that require location-based optimization and automation. In this paper, we present CO-LEELM, a continuous-output location fingerprinting method that combines two existing location fingerprinting methods to produce better accuracy in a dynamic environment where training data and reference devices are sparsely-distributed. The proposed method incorporates the use of Extreme Learning Machine (ELM) to improve the training speed which is a crucial factor that affects the scalability of the method.

Evolutionary algorithms (EAs) are usually required to solve problems based on domination relationship among solutions. Often, the domination relationship is almost the sole source of knowledge that EAs can utilize, especially when the problem solving engine concerned is taken as a black box. In this paper, the domination landscape (DL), onto which an optimization problem (OP) can be mapped, is introduced. A DL may correspond to a cluster of OPs, implying that a class of OPs may have the same DL. To illustrate DL, we consider its representation as a directed graph, with its corresponding matrix and function. Of the various properties of DL, the domination-preserving property is used for the analysis of DL-equivalent OPs, and for the basis for classification of OPs. Taking DL as a tool for theoretical analysis, parameters determination for fitness scaling, the convergence property of EAs and the analysis of robustness in light of fitness noise are presented. The study of DL in this paper establishes the necessary theoretical foundation for future applications of DL equality and similarity based optimization.

The use of robotics and autonomous systems to perform various missions has become increasingly popular and widely accepted. The scope of missions that can be performed by such systems includes complex tasks that require coordination and cooperation between multiple robots, not just simple repetitive tasks. Mission planner plays a vital role in making sure that the tasks performed by the robots are conducted in an efficient manner and meet all the mission requirements. In designing a mission planner, it is crucial to consider not only the functional aspect of the system but also its modularity, scalability, and interoperability. In this paper, we proposed an architecture of a heterogeneous multi-robot mission planner system that is based on robotic operating system (ROS). The system is designed to support heterogeneous multi-robot, where there are several types of robots having unique sets of capabilities and properties, which contributes greatly to the complexity of the mission planner.

From a computing perspective, a meme denotes information that represents knowledge, patterns, rules, or strategies used to solve complex problems. When applied on a problem, memes help a solver to arrive at good quality solutions more efficiently, guiding the search process according to certain procedures or rules, instead of randomly searching through the solution space. Depending on the complexity of the problems, evaluating the suitability of memes and selecting a set of effective memes for different problems, however, are not straightforward tasks. A meme that works well for some problems may not be effective for other problems. Besides, different memes might have different degrees of importance in solving a problem. The level of importance of each meme might also change at different stages of the search. In this paper, we discuss how multiple memes can be generated and applied to solve computational optimization problems. A case study in combinatorial optimization is also presented and discussed.

Reacting to situational needs for disaster relief measures and deployment of limited resources is a two-pronged operational and procedural protocol. It involves pre-disaster scenario-based planning and subsequently real-time on-the-field execution. In view of the potential complexity in terms of scale and magnitude involved, the pre-disaster stage has taken on a greater significance in an effort to achieve a state of operational readiness. In this article, we outline OptiSmart, a fully integrative planning system that can help fulfill such requirements. The system incorporates state-of-the-art optimization algorithms to carry out planning and scheduling, coupled with effective visualization and simulation tools to facilitate high-level operational support planning. This is applicable to both pre-disaster scenario planning as well as on the ground deployment and execution of real-time disaster management of operational protocol.

With the rapid growing market of wireless devices, positioning systems that make use of the signal strength of wireless devices are gaining more interest nowadays. Being able to track the location of a Wi-Fi or Radio Frequency Identification device could improve the quality of services in various sectors, including security, warehouse, logistic management, and healthcare. As compared with outdoor environment, positioning systems face a greater challenge in indoor environment because wireless signal is significantly influenced by building layout and surrounding objects, for which a location fingerprinting approach is needed. Moreover, the signal strength of a wireless device may also change over time, which is known as temporal variation, and therefore a reliable location estimation system must have the ability to learn and adapt with temporal changes. However, if the learning process is highly complex and requires long processing time, deploying the system into a larger scale would not be feasible. In recent years, Extreme Learning Machine (ELM) has surfaced as a viable alternative that challenged the norm of iterative and progressive learning. ELM has also been considered as a solution for indoor location fingerprinting. However, there has not been a comprehensive review on how the ELM-based approaches are linked with existing location fingerprinting techniques. Here we discuss some major location fingerprinting techniques, which are nearest-neighbor, LANDMARC, and LEMT, and formulate a new framework for systematically translating the techniques into ELM-based methods.

Indoor positioning is a key technology that enables the development of various smart and autonomous indoor applications. In recent years, wireless device is considered a promising technology enabler for indoor positioning despite facing various problems in the form of spatial, temporal, and device variation. Here we present a survey of problems in wireless-based indoor positioning and the state-of-the-art approaches aimed to solving those problems. This survey paper discusses various positioning metrics, parametric and non-parametric models to deal with spatial and temporal variation, as well as various methods that address device heterogeneity. In addition, some interesting fusion approaches are discussed. This paper aims to help researchers understand the current state of progress in wireless-based indoor positioning and identify areas that still need more research and improvement.

Forecasting electricity prices has been a widely investigated research issue in the deregulated power market scenario. High price volatilities, price spikes caused by a number of factors such as weather uncertainty, fluctuating fuel prices, transmission bottlenecks, etc., make the task of accurate price forecasting a formidable challenge for the market participants. A number of models have been proposed by researchers; however, achieving high accuracy is always not possible. In some specific applications such as self-scheduling by demand side participants, certain price thresholds are more useful than accurate price forecasts. In this paper, we have investigated the application of a novel neural network-based technique called extreme learning machine for the problem of classification of future electricity prices with respect to certain price thresholds. Different models corresponding to different lead times are developed and tested with data corresponding to Ontario and PJM markets. It is observed that classification with ELM is fast, less sensitive to user defined parameters and easily implementable.

Band selection plays an important role in identifying the most useful and valuable information contained in the hyperspectral images for further data analysis such as classification, clustering, etc. Memetic algorithm (MA), among other metaheuristic search methods, has been shown to achieve competitive performances in solving the NP-hard band selection problem. In this paper, we propose a formal probabilistic memetic algorithm for band selection, which is able to adaptively control the degree of global exploration against local exploitation as the search progresses. To verify the effectiveness of the proposed probabilistic mechanism, empirical studies conducted on five well-known hyperspectral images against two recently proposed state-of-the-art MAs for band selection are presented.

Wind power is becoming increasingly popular as a renewable source of energy. Being a non-dispatchable energy resource, wind power facilities entail efficient forecast mechanisms to estimate the production of various wind power utilities available. In an integrated grid system, a balance must be maintained between production and consumption. Given that wind power is directly affected by meteorological factors (wind speed etc.) accurately predicting such fluctuations becomes extremely important. These events of fluctuation are termed as ramp events. Forecast of wind power is important but accurate prediction of ramp events is much more crucial to the safety of the grid as well as the security and reliability of the grid. In this paper we employ the ELM (Extreme Learning Machine) technique on wind power data of 2012 Alberta, Canada market for different sampling times to predict wind power ramp events. We also try to compare it with respect to other existing standard algorithms of feed-forward Neural Networks to analyze the efficacy of the technique in the area. ELM is shown to outperform other techniques in terms of computation time whereas prediction performance is at par with other neural network algorithms.

Signal strength can be used to estimate location of a wireless device. As compared to other signal measures such as time-based and angle-based metrics, signal strength is normally embedded in wireless transceivers. This allows us to add location estimation feature on top of any wireless systems without requiring hardware modification. However, signal strength is affected by many environmental factors which cause temporal and spatial variation that could degrade the accuracy of location estimation system if not handled properly. In this paper, we focus on the temporal variation effect which is inevitable in dynamic environments where people and surrounding objects are typically not stationary. We try to improve the Location Estimation using Model Trees (LEMT) algorithm, a previous work that uses M5 model tree, by proposing that the calibration of the radio map over time can be done using Equality Constrained-optimization-based Semi-Supervised Extreme Learning Machine (ECSS-ELM). By using continuous signal strength readings collected from reference tags and tracking tag of a 2.4-GHz Radio Frequency Identification (RFID) system, we found that the algorithm can achieve comparable performance with much faster training time and testing time as compared to the M5 model tree.

In recent decades, a plethora of dedicated evolutionary algorithms (EAs) have been crafted to solve domain-specific complex problems more efficiently. Many advanced EAs have relied on the incorporation of domain-specific knowledge as inductive biases that is deemed to fit the problem of interest well. As such, the embedment of domain knowledge about the underlying problem within the search algorithms is becoming an established mode of enhancing evolutionary search performance. In this paper, we present a study on evolutionary memetic computing paradigm that is capable of learning and evolving knowledge meme that traverses different but related problem domains, for greater search efficiency. Focusing on combinatorial optimization as the area of study, a realization of the proposed approach is investigated on two NP-hard problem domains (i.e., capacitated vehicle routing problem and capacitated arc routing problem). Empirical studies on well-established routing problems and their respective state-of-the-art optimization solvers are presented to study the potential benefits of leveraging knowledge memes that are learned from different but related problem domains on future evolutionary search.

The advancement in game technology has served to enrich player’s gaming experience in a substantial way. Nowadays, it is common to have blockbuster quality games, with realistic graphics and engaging stories. Despite this, the progress made in incorporating Artificial Intelligence has been slow, and realistic human-like intelligence in games is hardly to be found. There have been some attempts to use Machine Learning in games, but such attempts often ended up impractical or affecting the players enjoyment due to several constraining factors. In this paper, we describe Meme War as a proof-of-concept for practical usage of Machine Learning in games. We introduce Extreme Learning Machine (ELM) as one approach to achieve a better experience in employing Machine Learning in games. Advantages of ELM over Multilayer Perceptron (MLP) are presented in terms of what ELM can offer in a practical point of view of the player.

For many years, various neural network models have been used to solve regression, binary classification, and multi-class classification problems. Their performance has been extensively compared against each other in terms of testing accuracy and training time. For multi-class classification problem, testing accuracy does not always give comprehensive information about the performance. It only shows the number of false detections without any clues on the false detection distribution. In this work we propose a cross-validation based on Extreme Learning Machine to identify classes that are found to have high number of false detections. These classes are treated as indistinguishable classes that need further processing or information. Our simulation shows that our proposed method is able to detect indistinguishable classes in three data sets. We also found that when indistinguishable classes exist, the training accuracy can be higher if each pair of those classes are marked as one merged class.