Search

Entrepreneurs and small businesses have increasingly used reward-based crowdfunding to raise capital for their creative projects, whose success is central to this industry. Thus, predicting the success of crowdfunding campaigns is a topic of great importance for entrepreneurs and platform managers. The literature that employs monolithic classifiers and static ensemble learning for crowdfunding success prediction are scarce. In contrast, the dynamic selection (DS) algorithm, which belongs to the ensemble learning category, deserves a particular remark since it has overcome traditional monolithic classifiers and static ensembles in many applications. This paper proposes a dynamic selection framework for reward crowdfunding prediction. DS algorithms select a competent subset of the classifier per query instance. This procedure is performed during the generalization, and the subset is composed of local experts, favoring an increase in accuracy. Fifteen machine learning models are evaluated using three metrics (accuracy, area under the ROC curve and F-score), and ensemble learning obtained better results than traditional classifiers. In particular, Meta-DES, which performs dynamic selection, obtains the best overall results among the evaluated models. Furthermore, since usually interpreting the output of ML models is considered to be very difficult due to their complex "black box" architecture, we also use Shapley additive explanations to interpret the perdition's outputs. Among variables evaluated in our models, the textual sentiment of the mass media, the number of pledges, and the target amount of the campaign deserve a highlight when predicting the campaign's success. The source-code and further details about the experimental analysis are available at https://github.com/las33/Crowdfunding. [ABSTRACT FROM AUTHOR]

Recently, learning classifier systems (LCS) have been used for a variety of Internet of Things (IoT) devices and multi-cloud services, including cloud-based centralized control of physical robots and actuators in continuous-time environments. Performance analysis of sensors, navigation of humanoid robots and intelligent control of rescue systems. In these systems, we can run evolutionary or intuitive algorithms on cloud servers to search the space of rules and simultaneously other learning processes to assign how to interact with the environment to the rules in the classification. Also, the problem of continuous congestion, traffic reduction, network routing and predicting traffic conditions in wireless networks is the main challenge facing these systems in real environments. Usually such systems are non-Markovian. Therefore, they need memory to save system states. This paper presents a framework for XCS-based memory-based LCS. In addition to identifying optimal rules in overlapping modes, the XCS architecture is equipped with a memory. Memory stores the most efficient classifier rules. These rules reduce the steps to reach the goal. In the first proposed method, only those rules that affect the moving motion are kept in memory. As the number of rules increases, some of them are deleted according to memory space. In the second proposed method, some features are added to the rules of this memory and the performance of the memory is optimized using evolutionary algorithms, which are used to remove the less used rules. The relative success of the proposed LCS architecture in solving well-known maze problems compared to the conventional XCS architecture confirms its efficiency in increasing the number of successes and reducing the steps to reach the goals. The results of this research are suggested as a suitable solution for reducing routing time, reducing network load in the problem of congestion and traffic in solving problems related to wireless networks. [ABSTRACT FROM AUTHOR]

This paper describes and evaluates the performance of a learning classifier system (lcs) inspired algorithm called Temporal Reinforcement And Classification Architecture (traca) on maze navigation tasks which contain hidden state. The evaluation of traca includes comparisons with other learning algorithms on selected difficult maze navigation tasks. Not all lcss are capable of learning all types of hidden-state mazes so traca is specifically compared against selected other lcs-based approaches that are most capable on these tasks, including xcsmh, AgentP (G), and AgentP (SA). Each algorithm is evaluated using a maze navigation task that has been identified as among the most difficult due to recurring aliased regions. The comparisons between algorithms include training time, test performance, and the size of the learned rule sets. The results indicate that each algorithm has its own advantages and drawbacks. For example, on the most difficult maze traca's average steps to the goal are 10.1 while AgentP (G) are 7.87; however, traca requires an average of only 354 training trials compared with 537 for AgentP (G). Following the maze tasks, traca is also tested on two variations in a truck driving task where it must learn to navigate four lanes of slower vehicles while avoiding collisions. The results show that traca can achieve a low number of collisions with relatively few trials (as low as 24 collisions over 5000 time steps after 10,000 training time steps) but may require multiple network construction attempts to achieve high performance. [ABSTRACT FROM AUTHOR]

As a genetics-based machine learning technique, zeroth-level classifier system (ZCS) based on average reward reinforcement learning (ZCSAR) evolves solutions to optimize average reward per time step. However, initial experimental results have shown that, in some cases, the performance of ZCSAR oscillates heavily during the learning period, or cannot reach the optimum during the testing period. In this paper, we modify the selection strategies in ZCSAR to improve its performance, under conditions of minimal changes of ZCSAR. The proposed selection strategies take tournament selection method to choose parents in Genetic Algorithm (GA), and take roulette wheel selection method to choose actions in match set and to choose classifiers for deletion in both GA and covering. Experimental results show that ZCSAR with the new selection strategies can evolve more promising solutions with enough parameter independence, and also with slighter oscillation during the learning period. [ABSTRACT FROM AUTHOR]

An ensemble of classifiers combines several single classifiers to deliver a final prediction or classification decision. An increasingly provoking question is whether such an ensemble can outperform the single best classifier. If so, what form of ensemble learning system (also known as multiple classifier learning systems) yields the most significant benefits in the size or diversity of the ensemble? In this paper, the ability of ensemble learning to predict and identify factors that influence or contribute to autism spectrum disorder therapy (ASDT) for intervention purposes is investigated. Given that most interventions are typically short-term in nature, henceforth, developing a robotic system that will provide the best outcome and measurement of ASDT therapy has never been so critical. In this paper, the performance of five single classifiers against several multiple classifier learning systems in exploring and predicting ASDT is investigated using a dataset of behavioural data and robot-enhanced therapy against standard human treatment based on 3000 sessions and 300 h, recorded from 61 autistic children. Experimental results show statistically significant differences in performance among the single classifiers for ASDT prediction with decision trees as the more accurate classifier. The results further show multiple classifier learning systems (MCLS) achieving better performance for ASDT prediction (especially those ensembles with three core classifiers). Additionally, the results show bagging and boosting ensemble learning as robust when predicting ASDT with multi-stage design as the most dominant architecture. It also appears that eye contact and social interaction are the most critical contributing factors to the ASDT problem among children. [ABSTRACT FROM AUTHOR]

Peroxidase mimicking Fe3O4@Chitosan (Fe3O4@Chi) nanozyme was synthesized and used for high-sensitive enzyme-free colorimetric detection of H2O2. The nanozyme was characterized in comparison with Fe3O4 nanoparticles (NPs) using X-ray diffraction, Fourier-transform infrared spectroscopy, dynamic light scattering, and thermogravimetric analysis. The catalytic performance of Fe3O4@Chi nanozyme was first evaluated by UV–Vis spectroscopy using 3,3′,5,5′-tetramethylbenzidine. Unlike Fe3O4NPs, Fe3O4@Chi nanozyme exhibited an intrinsic peroxidase activity with a detection limit of 69 nM. Next, the nanozyme was applied to a microfluidic paper-based analytical device (µPAD) and colorimetric analysis was performed at varying concentrations of H2O2 using a machine learning-based smartphone app called "Hi-perox Sens++." The app with machine learning classifiers made the system user-friendly as well as more robust and adaptive against variation in illumination and camera optics. In order to train various machine learning classifiers, the images of the µPADs were taken at 30 s and 10 min by four smartphone brands under seven different illuminations. According to the results, linear discriminant analysis exhibited the highest classification accuracy (98.7%) with phone-independent repeatability at t = 30 s and the accuracy was preserved for 10 min. The proposed system also showed excellent selectivity in the presence of various interfering molecules and good detection performance in tap water. [ABSTRACT FROM AUTHOR]

It is possible to model trust as an investment game, where a player in order to receive a reward or a better outcome, accepts a certain risk of defection by another player. Despite having achieved interesting insights and conclusions, traditional game theory does not predict the existence of trust between players who are selfish and exhibit maximizing behavior. However, experiments with these games reveal the presence of trust in player decision making. The purpose of this paper is twofold. First, it aims to build an agent-based economic model to show that trust revealed in these experiments can emerge from a simple set of dynamics. Using the generative methodology proposed by Epstein, we introduce natural selection, learning and group formation to the model to verify their impact on the emergence of trust between agents. Second, since the experiments reveal that the participants present bounded rational behavior, the paper aims to show that in an agent-based model, bounded rationality can be modelled through an artificial intelligence algorithm, the learning classifier system (LCS). As a result, we have observed that natural selection favors more selfish behavior. In addition, learning and the forming of groups increased trust in our simulations and they were able to reverse selfish behavior when introduced along with natural selection. The level of trust that emerged from the model with these three dynamics was similar to that observed in these experiments. Finally, it is possible to verify that the LCS was able to model bounded rational behavior in agents. [ABSTRACT FROM AUTHOR]

Learning Classifier Systems (LCSs) are a group of rule-based evolutionary computation techniques, which have been frequently applied to data mining tasks. The LCSs' rules are designed to be human-readable to enable the underlying knowledge to be investigated. However, the models for the majority of domains with high feature interaction contain a large number of rules that cooperatively represent the knowledge. However, the interaction between many rules is too complex to be comprehended by humans. Thus, it is hypothesized that translating the models' underlying patterns into human-discernable visualizations will advance the understanding of the learned patterns and LCSs themselves. Interrogatable artificial Boolean domains with varying numbers of attributes are considered as benchmarks. Three new visualization techniques, termed as Feature Importance Map, Action-based Feature Importance Map, and Action-based Feature's Average value Map, successfully produce interpretable results for all the complex domains tested. This includes both tracing the training progress and analyzing the trained models from LCSs. The visualization techniques' ability to handle complex optimal solutions is observed for the 14-bits Majority-On problem, where the patterns from 6435 different cooperating rules were translated into human-discernable graphs. [ABSTRACT FROM AUTHOR]

Underwater image classification is a challenging task because of challenging underwater environment and lighting conditions. We propose a novel learning classifier system (LCS), which can classify large-size underwater images with promising accuracy, and acquire knowledge in interpretable format, by using a novel classification convolution autoencoder (CCAE). In proposed system, CCAE is designed as a hybrid network, which combines benefits of classification and autoencoder, to extract compressed non-trivial features. It is also used to decompress LCS generated rules to original input space. In order to evaluate effectiveness of proposed solution, experiments are conducted on selected underwater synsets of benchmark ImageNet dataset. Results are compared with famous CNN methods based on parameters such as accuracy, precision, recall and F-measure. Experiments show that proposed system can accurately classify large-size underwater images with promising accuracy and outperforms well-known deep CNN methods. It has also been observed that LCS generated rules are well generalized, accurate and interpretable. [ABSTRACT FROM AUTHOR]

Predicting if a set of mushrooms is edible or not corresponds to the task of classifying them into two groups—edible or poisonous—on the basis of a classification rule. To support this binary task, we have collected the largest and most comprehensive attribute based data available. In this work, we detail the creation, curation and simulation of a data set for binary classification. Thanks to natural language processing, the primary data are based on a text book for mushroom identification and contain 173 species from 23 families. While the secondary data comprise simulated or hypothetical entries that are structurally comparable to the 1987 data, it serves as pilot data for classification tasks. We evaluated different machine learning algorithms, namely, naive Bayes, logistic regression, and linear discriminant analysis (LDA), and random forests (RF). We found that the RF provided the best results with a five-fold Cross-Validation accuracy and F2-score of 1.0 ( μ = 1 , σ = 0 ), respectively. The results of our pilot are conclusive and indicate that our data were not linearly separable. Unlike the 1987 data which showed good results using a linear decision boundary with the LDA. Our data set contains 23 families and is the largest available. We further provide a fully reproducible workflow and provide the data under the FAIR principles. [ABSTRACT FROM AUTHOR]

Human beings follow a continuous learning paradigm, i.e., they learn to solve smaller and relatively easy problems, retain the learnt knowledge and apply that knowledge to learn and solve more complex and large-scale problems of the domain. Currently, most machine learning and evolutionary computing systems lack this ability to reuse the previous learnt knowledge. This paper presents a lifelong machine learning model for text classification that extracts the useful knowledge from simple problems of a domain and reuses the learnt knowledge to learn complex problems of the domain. The proposed approach adopts a rule-based learning classifier system, and a rich encoding scheme is used to extract and reuse building units of knowledge. The experimental results show that the continuous learning approach outperformed the baseline classifier system. [ABSTRACT FROM AUTHOR]

Online structure learning approaches, such as those stemming from statistical relational learning, enable the discovery of complex relations in noisy data streams. However, these methods assume the existence of fully-labelled training data, which is unrealistic for most real-world applications. We present a novel approach for completing the supervision of a semi-supervised structure learning task. We incorporate graph-cut minimisation, a technique that derives labels for unlabelled data, based on their distance to their labelled counterparts. In order to adapt graph-cut minimisation to first order logic, we employ a suitable structural distance for measuring the distance between sets of logical atoms. The labelling process is achieved online (single-pass) by means of a caching mechanism and the Hoeffding bound, a statistical tool to approximate globally-optimal decisions from locally-optimal ones. We evaluate our approach on the task of composite event recognition by using a benchmark dataset for human activity recognition, as well as a real dataset for maritime monitoring. The evaluation suggests that our approach can effectively complete the missing labels and eventually, improve the accuracy of the underlying structure learning system. [ABSTRACT FROM AUTHOR]

Learning classifier systems are leading evolutionary machine learning systems that employ genetic algorithms to search for a set of optimally general and correct classification rules for a variety of machine learning problems, including supervised classification data mining problems. The curse of dimensionality is a phenomenon that arises when analysing data in high-dimensional spaces. Performance issues when dealing with increasing dimensionality in the training data, such as poor classification accuracy and stalled genetic search, are well known for learning classifier systems. However, a systematic study to establish the relationship between increasing dimensionality and learning challenges in these systems is lacking. The aim of this paper is to analyse the behaviour of Michigan-style learning classifier systems that use the most commonly adopted and expressive interval-based rules representation, under curse of dimensionality (also known as Hughes Phenomenon) problem. In this paper, we use well-established and mathematically founded formal geometrical properties of high-dimensional data spaces and generalisation theory of these systems to propose a formulation of such relationship. The proposed formulations are validated experimentally using a set of synthetic, two-class classification problems. The findings demonstrate that the curse of dimensionality occurs for as few as ten dimensions and negatively affects the evolutionary search with a hyper-rectangular rule representation. A number of approaches to overcome some of the difficulties uncovered by the presented analysis are then discussed. Three approaches are then analysed in more detail using a set of synthetic, two-class classification problems. Experimental study demonstrates the effectiveness of these approaches to handle increasing dimensional data. [ABSTRACT FROM AUTHOR]

Reinforcement learning has great potential in solving practical problems, but when combining it with neural networks to solve small scale discrete space problems, it may easily trap in a local minimum value. Traditional reinforcement learning utilizes continuous updating of a single agent to learn policies, which easily leads to a slow convergence speed. In order to solve the above problems, we combine asynchronous methods with existing tabular reinforcement learning algorithms, propose a parallel architecture to solve the discrete space path planning problem, and present some new variants of asynchronous reinforcement learning algorithms. We apply these algorithms on the standard reinforcement learning environment problems, and the experimental results show that these methods can solve discrete space path planning problems efficiently. One of these algorithms, Asynchronous Phased Dyna-Q, which surpasses existing asynchronous reinforcement learning algorithms, can well balance exploration and exploitation. [ABSTRACT FROM AUTHOR]

Sentiment analysis of public views and spam detection from social media text messages are two challenging data analysis tasks due to short informal text. This paper investigates the performance of learning classifier systems (LCS), which are rule-based machine learning techniques, in sentiment analysis of twitter messages and movie reviews, and spam detection from SMS and email data sets. In this study, an existing LCS technique is extended by introducing a novel encoding scheme to represent classifier rules in order to handle the sparseness in feature vectors, which are generated using the term frequency inverse document frequency of word n-grams and sentiment lexicons. The obtained results show that the proposed encoding scheme smoothed the learning process and generated consistently good results in all experiments conducted in this study. [ABSTRACT FROM AUTHOR]

Because Particle Filter algorithm in video tracking has some problems, such as weight degradation, poor samples after resampling, being difficult to select the optimal important probability density, being difficult to accurately follow to the target again after tracking failure, the random forest learning algorithm is introduced into the Particle Filter tracking algorithm. The current estimated target state image is transmitted to the random forest detection module, the algorithm adopts reliable sample update strategy, the random forest model is used to detect the target area, and the decision tree detection module rapidly detects target state adjacent areas. In the case of target temporarily lost, two methods of collaboration eventually accurately allocate to the tracking target. The algorithm ensures the robustness of tracking, effectively avoids the drift problem. [ABSTRACT FROM AUTHOR]

This paper constructs a project portfolio selection model from the strategic perspective. Two goals are proposed for the portfolio to achieve, i.e., strategic contributions and financial returns. The uncertainties involved are addressed with fuzzy real options. Then, a modified multi-objective genetic algorithm is designed to determine the portfolios. Finally, a real case is provided to validate the model’s effectiveness. The results demonstrate that the proposed algorithm can optimize two objectives simultaneously and keep the plausible Pareto-optimal set which wins over the single-objective model solutions in achieving the shared value. [ABSTRACT FROM AUTHOR]

The research of automatic face recognition has attracted much attention from many researchers because of human faces’ uniqueness and usability. However, in the real-world applications, the acquisition equipment of face images is affected by illumination changes, facial expression variations, different postures and other environment factors, resulting in limited number of face images collected. This situation has become an obstacle to the development of face recognition technology. Therefore, in this paper, we utilize the information of the left-half face and right-half face to generate respectively two virtual ‘axis-symmetrical’ face images from an original face image and adopt collaborative representation based classification method (CRC) to perform classification. The first and second virtual face images convey more information of the right-half face and left-half face, respectively. Experiments have been performed on the Extended Yale_B, ORL, AR and FERET face databases and the experimental results show that our method can improve the recognition accuracy effectively. [ABSTRACT FROM AUTHOR]

Ensemble learning is a powerful approach for achieving more accurate predictions compared with single classifier. However, this powerful classification ability is achieved at the expense of heavy storage requirements and computational burdens on the ensemble. Ensemble pruning is a crucial step for the reduction of the predictive overhead without worsening the performance of original ensemble. This paper suggests an efficient and effective ordering-based ensemble pruning based on the induction of decision tree. The suggested method maps the dataset and base classifiers to a new dataset where the ensemble pruning can be transformed to a feature selection problem. Furthermore, a set of accurate, diverse and complementary base classifiers can be selected by the induction of decision tree. Moreover, an evaluation function that deliberately favors the candidate sub-ensembles with an improved performance in classifying low margin instances has also been designed. The comparative experiments on 24 benchmark datasets demonstrate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

A swap sequence-based particle swarm optimization (SSPSO) technique and genetic algorithm (GA) are used in tandem to develop a hybrid algorithm to solve generalized traveling salesman problem. Local search algorithm K-Opt is occasionally used to move any stagnant solution. Here, SSPSO is used to find the sequence of groups of a solution in which a tour to be made and cities from different groups of the sequence are selected using GA. The K-Opt algorithm (for K=3) is used periodically for a predefined number of iterations to improve the quality of the solutions. The algorithm is capable of solving the problem in crisp as well as in imprecise environment. For this purpose, a general fitness evaluation rule for the solutions is proposed. The efficiency of the algorithm is tested in crisp environment using different size benchmark problems available in TSPLIB. In crisp environment, the algorithm gives 100% success rate for problems up to considerably large sizes. Imprecise problems are generated from crisp problems randomly using a rule and are solved using the proposed approach. The obtained results are discussed. Moreover it is observed that the proposed algorithm finds multiple optimal paths, when they exist, both for the crisp problems and their fuzzy variations. [ABSTRACT FROM AUTHOR]

In this study, the multi-objective optimal design of hybrid viscoelastic/composite sandwich beams for minimum weight and minimum vibration response is aimed. The equation of motion for linear vibrations of a multi-layer beam is derived by using the principle of virtual work in the most general form. These governing equations together with the boundary conditions are discretized by the generalized differential quadrature method (GDQM) in the frequency domain for the first time. Also, the time and temperature dependent properties of the viscoelastic materials are taken into consideration by a novel ten-parameter fractional derivative model that can realistically capture the response of these materials. The material variability is accounted for by letting an optimization algorithm choose a material freely out of four fiber-reinforced composite materials and five viscoelastic damping polymers for each layer. The design parameters, i.e., the orientation angles of the composites, layer thicknesses and the layer materials that give the set of optimal solutions, namely the Pareto frontier, is obtained for the three and nine-layered clamped-free sandwich beams by using a variant of the non-dominated sorting genetic algorithms (NSGA II). [ABSTRACT FROM AUTHOR]

Supplementary information, such as multi-temporal spectral data and textural features, has the potential to improve land cover classification accuracy. However, given the larger volumes of remote sensing data, it is difficult to utilize all the features of remote sensing big data having different times and spatial resolutions. Inefficiency is also a large problem when dealing with large area land cover mapping. In this study, a new mode of incorporating spatial and temporal dependencies in a complex region employing the random forests (RFs) classifier was utilized. To map land covers, spring and autumn spectral images and their spectral indexes, textural features obtained from Landsat 5 were selected, and an importance measure variable was used to reduce the data's dimension. In addition to randomly selecting the variable, we used random sampling to furthest decrease the generalization error in RF. The results showed that utilizing random sampling, multi-temporal spectral image and texture features, the classification of the Wuhan urban agglomeration, China, using RF performed well. The RF algorithm yielded an overall accuracy of 89.2% and a Kappa statistic of 0.8522, indicating high model performance. In addition, the variable importance measures demonstrated that the type of textural features was extremely important for intra-class separability. The RF model has transitivity. The algorithm can be extended by choosing a set of appropriate features for signature extension over large areas or in time-series of Landsat imagery. Land cover mapping might be more economical and efficient if no-cost imagery is used. [ABSTRACT FROM AUTHOR]

A novel evolving semi-supervised classifier, namely Parsimonious Classifier+ (pClass+), is proposed in this paper. pClass+ enhances a recently developed classifier, namely pClass, for a semi-supervised learning scenario. As with its predecessor, pClass+ is capable of initiating its learning process from scratch with an empty rule base and adopts an open network structure, where fuzzy rules are evolved, pruned, and recalled automatically on demands. The novelty of pClass+ lies in an online active learning technique, which decreases operator's annotation efforts and expedites its training process. pClass+ is also equipped with a new parameter identification strategy to cope with the class overlapping situation. The efficacy of pClass+ has been experimentally validated with numerous synthetic and real-world study cases, confirmed by thorough statistical tests and comparisons against state-of-the art classifiers, where pClass+ outperforms its counterparts in achieving the best trade-off between accuracy and complexity. [ABSTRACT FROM AUTHOR]

A unit hydrograph (UH) of a watershed may be viewed as the unit pulse response function of a linear system. In recent years, the use of probability distribution functions (pdfs) for determining a UH has received much attention. In this study, a nonlinear optimization model is developed to transmute a UH into a pdf. The potential of six popular pdfs, namely two-parameter gamma, two-parameter Gumbel, two-parameter log-normal, two-parameter normal, three-parameter Pearson distribution, and two-parameter Weibull is tested on data from the Lighvan catchment in Iran. The probability distribution parameters are determined using the nonlinear least squares optimization method in two ways: (1) optimization by programming in Mathematica; and (2) optimization by applying genetic algorithm. The results are compared with those obtained by the traditional linear least squares method. The results show comparable capability and performance of two nonlinear methods. The gamma and Pearson distributions are the most successful models in preserving the rising and recession limbs of the unit hydographs. The log-normal distribution has a high ability in predicting both the peak flow and time to peak of the unit hydrograph. The nonlinear optimization method does not outperform the linear least squares method in determining the UH (especially for excess rainfall of one pulse), but is comparable. [ABSTRACT FROM AUTHOR]

There is no uniform approach in the literature for modelling sequential correlations in sequence classification problems. It is easy to find examples of unstructured models ( e.g. logistic regression) where correlations are not taken into account at all, but there are also many examples where the correlations are explicitly incorporated into a-potentially computationally expensive-structured classification model ( e.g. conditional random fields). In this paper we lay theoretical and empirical foundations for clarifying the types of problem which necessitate direct modelling of correlations in sequences, and the types of problem where unstructured models that capture sequential aspects solely through features are sufficient. The theoretical work in this paper shows that the rate of decay of auto-correlations within a sequence is related to the excess classification risk that is incurred by ignoring the structural aspect of the data. This is an intuitively appealing result, demonstrating the intimate link between the auto-correlations and excess classification risk. Drawing directly on this theory, we develop well-founded visual analytics tools that can be applied a priori on data sequences and we demonstrate how these tools can guide practitioners in specifying feature representations based on auto-correlation profiles. Empirical analysis is performed on three sequential datasets. With baseline feature templates, structured and unstructured models achieve similar performance, indicating no initial preference for either model. We then apply the visual analytics tools to the datasets, and show that classification performance in all cases is improved over baseline results when our tools are involved in defining feature representations. [ABSTRACT FROM AUTHOR]

We provide a unifying perspective for two decades of work on cost-sensitive Boosting algorithms. When analyzing the literature 1997-2016, we find 15 distinct cost-sensitive variants of the original algorithm; each of these has its own motivation and claims to superiority-so who should we believe? In this work we critique the Boosting literature using four theoretical frameworks: Bayesian decision theory, the functional gradient descent view, margin theory, and probabilistic modelling. Our finding is that only three algorithms are fully supported-and the probabilistic model view suggests that all require their outputs to be calibrated for best performance. Experiments on 18 datasets across 21 degrees of imbalance support the hypothesis-showing that once calibrated, they perform equivalently, and outperform all others. Our final recommendation-based on simplicity, flexibility and performance-is to use the original Adaboost algorithm with a shifted decision threshold and calibrated probability estimates. [ABSTRACT FROM AUTHOR]

Dictionary learning plays a key role in image representation for classification. A multi-modal dictionary is usually learned from feature samples across different classes and shared in the feature encoding process. Ideally each atom in dictionary corresponds to a single class of images, while each class of images corresponds to a certain group of atoms. Image features are encoded as linear combinations of selected atoms in a given dictionary. We propose to use elastic net as regularizer to select atoms in feature coding and related dictionary learning process, which not only benefits from the sparsity similar as ℓ penalty but also encourages a grouping effect that helps improve image representation. Experimental results of image classification on benchmark datasets show that with dictionary learned in the proposed way outperforms state-of-the-art dictionary learning algorithms. [ABSTRACT FROM AUTHOR]

Learning classifier systems (LCS) are population-based reinforcement learners that were originally designed to model various cognitive phenomena. This paper presents an explicitly cognitive LCS by using spiking neural networks as classifiers, providing each classifier with a measure of temporal dynamism. We employ a constructivist model of growth of both neurons and synaptic connections, which permits a genetic algorithm to automatically evolve sufficiently-complex neural structures. The spiking classifiers are coupled with a temporally-sensitive reinforcement learning algorithm, which allows the system to perform temporal state decomposition by appropriately rewarding 'macro-actions', created by chaining together multiple atomic actions. The combination of temporal reinforcement learning and neural information processing is shown to outperform benchmark neural classifier systems, and successfully solve a robotic navigation task. [ABSTRACT FROM AUTHOR]

In this paper, we present effective algorithms to automatically annotate clothes from social media data, such as Facebook and Instagram. Clothing annotation can be informally stated as recognizing, as accurately as possible, the garment items appearing in the query photo. This task brings huge opportunities for recommender and e-commerce systems, such as capturing new fashion trends based on which clothes have been used more recently. It also poses interesting challenges for existing vision and recognition algorithms, such as distinguishing between similar but different types of clothes or identifying a pattern of a cloth even if it has different colors and shapes. We formulate the annotation task as a multi-label and multi-modal classification problem: (i) both image and textual content (i.e., tags about the image) are available for learning classifiers, (ii) the classifiers must recognize a set of labels (i.e., a set of garment items), and (iii) the decision on which labels to assign to the query photo comes from a set of instances that is used to build a function, which separates labels that should be assigned to the query photo, from those that should not be assigned. Using this configuration, we propose two approaches: (i) the pointwise one, called MMCA, which receives a single image as input, and (ii) a multi-instance classification, called M3CA, also known as pairwise approach, which uses pair of images to create the classifiers. We conducted a systematic evaluation of the proposed algorithms using everyday photos collected from two major fashion-related social media, namely and . Our results show that the proposed approaches provide improvements when compared to popular first choice multi-label, multi-modal, multi-instance algorithms that range from 20 % to 30 % in terms of accuracy. [ABSTRACT FROM AUTHOR]

Application of multiobjective optimization in sewerage rehabilitation management is not widespread due to the limitation of data collection and complex optimization process. Thus, a few researches in literature focused on sewerage rehabilitation optimization, and only considered two-objective optimization usually between the service life and the direct cost instead of a social cost. A sewerage rehabilitation multiobjective optimization decision support system (SRMOS) was developed for sewerage rehabilitation management in this study. The nondominated sorting genetic algorithm-II was used to design a set of Pareto surfaces with desirable rehabilitation effectiveness at the lowest cost by providing optimal plans comprising a construction method and substitute material. The SRMOS was applied to a real sewerage system to provide tradeoff solutions for three conflicting objectives, which are minimizing rehabilitation cost, maximizing pipe service, and minimizing traffic disruption. Compared with the experts' manual estimation, the plan derived from the SRMOS enables saving nearly 20 % of the rehabilitation cost. The contours of the rehabilitation cost show the equivalent relation between the traffic disruption and service life of pipes. The results indicate that increasing the number of objectives can make up the drawback of cost hard to be quantified and can also facilitate deriving practical plans for reference in decision-making. [ABSTRACT FROM AUTHOR]

This paper presents a method for synthesizing linear-phase FIR filters capable of implementing magnitude characteristics defined arbitrarily by the user through a set of frequency-magnitude points, filters that are optimized with respect to several criteria. The main idea is to approach the filter synthesis as a multi-objective optimization problem, targeting the minimization of both the peak magnitude and the total squared errors of the resulting magnitude characteristics, as well as implementation-related requirements such as the reduction of the filter length. The optimization procedure uses a genetic algorithm tailored to this application; it employs a novel encoding scheme for the filter chromosome and an efficient fitness function, based on only two well-chosen constraints. Several design examples are presented: first, optimized synthesis of FIR filters with magnitude characteristics that match given (arbitrary) human audiograms, and second, synthesis of filters defined by parameters related to their pass- and stop-bands. The results yielded by the proposed method compare well with filters synthesized by means of previously reported methods and an industry-standard MATLAB tool. [ABSTRACT FROM AUTHOR]

The Three-Cornered Coevolution Framework describes a method that is capable of addressing classification tasks through coevolution (coadaptive evolution) where three different agents (i.e. a generation agent and two classification agents) learn and adapt to the changes of the problems without human involvement. Here, artificial problems can be generated in concert with classification agents in order to provide insight into their relationships. Previous work on the Two-Cornered Coevolution Framework provided foundation for implementing the system that was able to set-up the problem's difficulty appropriately while triggering the coevolutionary process. However, the triggering process was set manually without utilising the third agent as proposed in the original framework to perform this task. Previous work on the Three-Cornered Coevolution introduced the third agent (a new classification agent) to trigger the coevolutionary process within the system, where its functionality and effect on the system requires investigation. This paper details the implementation for this case; two classification agents that use different styles of learning techniques (e.g. supervised versus reinforcement learning techniques) is adapted in the classification agents to learn the various classification problems. Ultimately, Learning Classifier System (LCS) is chosen to be implemented in the participating agents. LCS has several potential characteristics, such as interpretability, generalization capability and variations in representation, that are suitable for the system. Experiments show that the Pittsburgh-style LCS with the adaptation of Tabu Search technique in S capable to autonomously adjust the problem's difficulty and generate a wide range of problems for classification. The adaptation of A-PLUS to an 'on-line' system is successful implemented. Further, the classification agents (i.e. R and I) are able to solve the classification tasks where the classification performance are varied. The Three-Cornered Coevolution Framework offers a great potential for autonomous learning and provides useful insight into coevolution learning over the standard studies of pattern recognition. The system is capable of autonomously generating various problems, learning and providing insight into each learning system's ability by determining the problem domains where they perform relatively well. This is in contrast to humans having to determine the problem domains. [ABSTRACT FROM AUTHOR]

Sequence labeling is an interesting classification domain where, like normal classification, every input has a class label, but unlike normal classification, prediction of an input's label may depend on the values of other inputs or their classes, and so a learner may need to refer to inputs and classes at different time stamps to classify the current input. This is more difficult because a learner does not know where and how many inputs are needed to classify the current input. Our interest is in learning general rules for sequence labeling. The XCS algorithm is a rule-based knowledge discovery system powered by a genetic algorithm which has often been used for classification. Here we present XCS-SL, an extension of XCS classifier system which can be applicable to sequence labeling. Towards an application of Learning Classifier System (LCS) to sequence labeling, we propose a new classifier condition with memory (called a variable-length condition) and a rule-discovery system for the new classifier condition, which enables XCS to apply it to sequence labeling. In XCS-SL, classification rules (called 'classifiers' here) can include extra conditions on previous inputs, which act as memories. In sequence labeling, the number of conditions/memories needed may be different for each input, hence, using a fixed number of conditions (i.e., fixed-length condition) for all classifiers is not a good solution. Instead, XCS-SL classifiers have a variable-length condition to provide more or less memory. The genetic algorithm can grow and shrink conditions to find a suitable memory size. On two synthetic benchmark problems XCS-SL learns optimal classifiers, and on a real-world sequence labeling task it derives high classification accuracy and discovers interesting knowledge that shows dependencies between inputs at different times. The comprehensively described system is the first application of a LCS to sequence labeling and we consider it a promising direction for future work. [ABSTRACT FROM AUTHOR]

Tile coding is an effective reinforcement learning method that uses a rather ingenious generalization mechanism based on (1) a carefully designed parameter setting and (2) the assumption that nearby states in the problem space will correspond to similar payoff predictions in the action-value function. Previously, we extended XCSF with tile coding prediction and compared it to tabular tile coding, showing that (1) XCSF performs as well as parameter-optimized tile coding, while also (2) evolving individualized parameter settings in each problem subspace. Our comparison was based on a set of well-known reinforcement learning environments (2D Gridworld and the Mountain Car) that involved no action-value discontinuities and so posed no challenge to tabular tile coding. In this paper, we go a step further and test XCSF with tile coding on a set of problems designed to challenge tile coding by introducing discontinuities in the action value landscape. The new testbed (called MazeWorld) extends 2D Gridworld with impenetrable obstacles, a conceptually simple modification that can dramatically increase the problem complexity for tabular tile coding. We compare four versions of XCSF with tile coding (each adapting a different set of parameters) to tabular tile coding on four problems of increasing complexity. We show that our system (1) needs fewer training problems than standard tile coding to reach an optimal policy; (2) it can evolve adequate coding parameters in each subspace without any previous knowledge; and that (3) even when XCSF is not allowed to evolve these parameters, the genetic algorithm will still adapt classifier conditions to properly decompose the problem into subspaces thus being much less sensitive to the parameter settings than tabular tile coding. [ABSTRACT FROM AUTHOR]

Predicting membrane protein type is a challenging problem, particularly when the query proteins may simultaneously have two or more different types. Most of the existing methods can only be used to deal with the single-label proteins. Actually, multiple-label proteins should not be ignored because they usually bear some special functions worthy of in-depth studies. By introducing the 'multi-labeled learning' and hybridizing evolution information through Grey-PSSM, a novel predictor called iMem-Seq is developed that can be used to deal with the systems containing both single and multiple types of membrane proteins. As a demonstration, the jackknife cross-validation was performed with iMem-Seq on a benchmark dataset of membrane proteins classified into the eight types, where some proteins belong to two or there types, but none has ≥25 % pairwise sequence identity to any other in a same subset. It was demonstrated via the rigorous cross-validations that the new predictor remarkably outperformed all its counterparts. As a user-friendly web-server, iMem-Seq is freely accessible to the public at the website . [ABSTRACT FROM AUTHOR]

Learning classifier systems (LCSs), an established evolutionary computation technique, are over 30 years old with much empirical testing and foundations of theoretical understanding. XCS is a well-tested LCS model that generates optimal (i.e., maximally general and accurate) classifier rules in the final solution. Previous work has hypothesized the evolution mechanisms in XCS by identifying the bounds of learning and population requirements. However, no work has shown exactly how an optimum rule is evolved or especially identifies whether the methods within an LCS are being utilized effectively. In this paper, we introduce a method to trace the evolution of classifier rules generated in an XCS-based classifier system. Specifically, we introduce the concept of a family tree, termed parent-tree, for each individual classifier rule generated in the system during training, which describes the whole generational process for that classifier. Experiments are conducted on two sample Boolean problem domains, i.e., multiplexer and count ones problems using two XCS-based systems, i.e., standard XCS and XCS with code-fragment actions. The analysis of parent-trees reveals, for the first time in XCS, that no matter how specific or general the initial classifiers are, all the optimal classifiers are converged through the mechanism 'be specific then generalize' near the final stages of evolution. Populations where the initial classifiers were slightly more specific than the known 'ideal' specificity in the target solutions evolve faster than either very specific, ideal or more general starting classifier populations. Consequently introducing the 'flip mutation' method and reverting the conventional wisdom back to apply rule discovery in the match set has demonstrated benefits in binary classification problems, which has implications in using XCS for knowledge discovery tasks. It is further concluded that XCS does not directly utilize all relevant information or all breeding strategies to evolve the optimum solution, indicating areas for performance and efficiency improvement in XCS-based systems. [ABSTRACT FROM AUTHOR]

In the family of learning classifier systems, the classifier system XCS has been successfully used for many applications. However, the standard XCS has no memory mechanism and can only learn optimal policy in Markov environments, but fails in non-Markov ones. In this work, we aim to develop a new classifier system based on XCS to tackle this problem. It adds a memory list with numbered slots to XCS to record input sensation history, and extends only a small number of classifiers with memory conditions. The classifier's memory condition, as a foothold to disambiguate non-Markov states, is used to sense a specified element in the memory list, which makes our system can 'jump over' irrelevant or confusing states to get decisive prior information that may be far back in time. Besides, a detection method is employed to recognize non-Markov states in environments, to avoid these states controlling over classifiers' memory conditions. Furthermore, four sets of different complex maze environments have been tested by the proposed method. Experimental results show that our system can overcome the overhead problem often encountered in history-window approaches, and is an effective technique to solve non-Markov environments. [ABSTRACT FROM AUTHOR]

Instance selection is an important research problem of data pre-processing in the data mining field. The aim of instance selection is to reduce the data size by filtering out noisy data, which may degrade the mining performance, from a given dataset. Genetic algorithms have presented an effective instance selection approach to improve the performance of data mining algorithms. However, current approaches only pursue the simplest evolutionary process based on the most reasonable and simplest rules. In this paper, we introduce a novel instance selection algorithm, namely a genetic-based biological algorithm (GBA). GBA fits a 'biological evolution' into the evolutionary process, where the most streamlined process also complies with the reasonable rules. In other words, after long-term evolution, organisms find the most efficient way to allocate resources and evolve. Consequently, we can closely simulate the natural evolution of an algorithm, such that the algorithm will be both efficient and effective. Our experiments are based on comparing GBA with five state-of-the-art algorithms over 50 different domain datasets from the UCI Machine Learning Repository. The experimental results demonstrate that GBA outperforms these baselines, providing the lowest classification error rate and the least storage requirement. Moreover, GBA is very computational efficient, which only requires slightly larger computational cost than GA. [ABSTRACT FROM AUTHOR]

We prove theoretical guarantees for an averaging-ensemble of randomly projected Fisher linear discriminant classifiers, focusing on the case when there are fewer training observations than data dimensions. The specific form and simplicity of this ensemble permits a direct and much more detailed analysis than existing generic tools in previous works. In particular, we are able to derive the exact form of the generalization error of our ensemble, conditional on the training set, and based on this we give theoretical guarantees which directly link the performance of the ensemble to that of the corresponding linear discriminant learned in the full data space. To the best of our knowledge these are the first theoretical results to prove such an explicit link for any classifier and classifier ensemble pair. Furthermore we show that the randomly projected ensemble is equivalent to implementing a sophisticated regularization scheme to the linear discriminant learned in the original data space and this prevents overfitting in conditions of small sample size where pseudo-inverse FLD learned in the data space is provably poor. Our ensemble is learned from a set of randomly projected representations of the original high dimensional data and therefore for this approach data can be collected, stored and processed in such a compressed form. We confirm our theoretical findings with experiments, and demonstrate the utility of our approach on several datasets from the bioinformatics domain and one very high dimensional dataset from the drug discovery domain, both settings in which fewer observations than dimensions are the norm. [ABSTRACT FROM AUTHOR]

We present new methods for multilabel classification, relying on ensemble learning on a collection of random output graphs imposed on the multilabel, and a kernel-based structured output learner as the base classifier. For ensemble learning, differences among the output graphs provide the required base classifier diversity and lead to improved performance in the increasing size of the ensemble. We study different methods of forming the ensemble prediction, including majority voting and two methods that perform inferences over the graph structures before or after combining the base models into the ensemble. We put forward a theoretical explanation of the behaviour of multilabel ensembles in terms of the diversity and coherence of microlabel predictions, generalizing previous work on single target ensembles. We compare our methods on a set of heterogeneous multilabel benchmark problems against the state-of-the-art machine learning approaches, including multilabel AdaBoost, convex multitask feature learning, as well as single target learning approaches represented by Bagging and SVM. In our experiments, the random graph ensembles are very competitive and robust, ranking first or second on most of the datasets. Overall, our results show that our proposed random graph ensembles are viable alternatives to flat multilabel and multitask learners. [ABSTRACT FROM AUTHOR]

We tackle the problem of learning linear classifiers from noisy datasets in a multiclass setting. The two-class version of this problem was studied a few years ago where the proposed approaches to combat the noise revolve around a Perceptron learning scheme fed with peculiar examples computed through a weighted average of points from the noisy training set. We propose to build upon these approaches and we introduce a new algorithm called Unconfused Multiclass additive Algorithm ( U MA) which may be seen as a generalization to the multiclass setting of the previous approaches. In order to characterize the noise we use the confusion matrix as a multiclass extension of the classification noise studied in the aforementioned literature. Theoretically well-founded, U MA furthermore displays very good empirical noise robustness, as evidenced by numerical simulations conducted on both synthetic and real data. [ABSTRACT FROM AUTHOR]

High reliability and low power consumption are among the major requirements in design of wireless sensor networks (WSNs). In this paper, a multi-objective problem is formulated as a Joint Power consumption and data Reliability (JPR) optimization problem. For this purpose, a connected dominating set (CDS)-based topology control approach is proposed. Our objective is to self-organize the network with minimum interference and power consumption. We consider the power changes into a topology with minimum CDS infrastructure subject to connectivity constraints. Since this problem is NP-hard, it cannot be dealt with using polynomial-time exact algorithms. Therefore, we first present a genetic algorithm taking into consideration problem-specific goals and constraints in an approximated manner called JPR Genetic Algorithm ( Jpr- ga). Secondly, a Hierarchical Sub-Chromosome Genetic Algorithm ( Hsc- ga) is proposed to obtain more accurate and faster solutions in the large and dense networks. We evaluate these algorithm over different networks topologies to analyse their efficiency. Comparing Jpr- ga and Hsc- ga with two different scenarios reveal that the proposed algorithms can efficiently balance power consumption and data communication reliability of sensor nodes and also prolong the network lifetime in WSNs. [ABSTRACT FROM AUTHOR]

In this paper, we show that the way internal estimates are used to measure variable importance in Random Forests are also applicable to feature selection in unsupervised learning. We propose a new method called Random Cluster Ensemble (RCE for short), that estimates the out-of-bag feature importance from an ensemble of partitions. Each partition is constructed using a different bootstrap sample and a random subset of the features. We provide empirical results on nineteen benchmark data sets indicating that RCE, boosted with a recursive feature elimination scheme (RFE) (Guyon and Elisseeff, Journal of Machine Learning Research, 3:1157-1182, ), can lead to significant improvement in terms of clustering accuracy, over several state-of-the-art supervised and unsupervised algorithms, with a very limited subset of features. The method shows promise to deal with very large domains. All results, datasets and algorithms are available on line (). [ABSTRACT FROM AUTHOR]