Lower Training Time Research Articles

High Class-Imbalance in pre-miRNA Prediction: A Novel Approach Based on deepSOM

The computational prediction of novel microRNA within a full genome involves identifying sequences having the highest chance of being a miRNA precursor (pre-miRNA). These sequences are usually named candidates to miRNA. The well-known pre-miRNAs are usually only a few in comparison to the hundreds of thousands of potential candidates to miRNA that have to be analyzed, which makes this task a high class-imbalance classification problem. The classical way of approaching it has been training a binary classifier in a supervised manner, using well-known pre-miRNAs as positive class and artificially defining the negative class. However, although the selection of positive labeled examples is straightforward, it is very difficult to build a set of negative examples in order to obtain a good set of training samples for a supervised method. In this work, we propose a novel and effective way of approaching this problem using machine learning, without the definition of negative examples. The proposal is based on clustering unlabeled sequences of a genome together with well-known miRNA precursors for the organism under study, which allows for the quick identification of the best candidates to miRNA as those sequences clustered with known precursors. Furthermore, we propose a deep model to overcome the problem of having very few positive class labels. They are always maintained in the deep levels as positive class while less likely pre-miRNA sequences are filtered level after level. Our approach has been compared with other methods for pre-miRNAs prediction in several species, showing effective predictivity of novel miRNAs. Additionally, we will show that our approach has a lower training time and allows for a better graphical navegability and interpretation of the results. A web-demo interface to try deepSOM is available at http://fich.unl.edu.ar/sinc/web-demo/deepsom/.

FP-ELM: An online sequential learning algorithm for dealing with concept drift

The online sequential extreme learning machine (OS-ELM) algorithm is an on-line and incremental learning method, which can learn data one-by-one or chunk-by-chunk with a fixed or varying chunk size. And OS-ELM achieves the same learning performance as ELM trained by the complete set of data. However, in on-line learning environments, the concepts to be learned may change with time, a feature referred to as concept drift. To use ELMs in such non-stationary environments, a forgetting parameters extreme learning machine (FP-ELM) is proposed in this paper. The proposed FP-ELM can achieve incremental and on-line learning, just like OS-ELM. Furthermore, FP-ELM will assign a forgetting parameter to the previous training data according to the current performance to adapt to possible changes after a new chunk comes. The regularized optimization method is used to avoid estimator windup. Performance comparisons between FP-ELM and two frequently used ensemble approaches are carried out on several regression and classification problems with concept drift. The experimental results show that FP-ELM produces comparable or better performance with lower training time.

Active learning based segmentation of Crohns disease from abdominal MRI

This paper proposes a novel active learning (AL) framework, and combines it with semi supervised learning (SSL) for segmenting Crohns disease (CD) tissues from abdominal magnetic resonance (MR) images. Robust fully supervised learning (FSL) based classifiers require lots of labeled data of different disease severities. Obtaining such data is time consuming and requires considerable expertise. SSL methods use a few labeled samples, and leverage the information from many unlabeled samples to train an accurate classifier. AL queries labels of most informative samples and maximizes gain from the labeling effort. Our primary contribution is in designing a query strategy that combines novel context information with classification uncertainty and feature similarity. Combining SSL and AL gives a robust segmentation method that: (1) optimally uses few labeled samples and many unlabeled samples; and (2) requires lower training time. Experimental results show our method achieves higher segmentation accuracy than FSL methods with fewer samples and reduced training effort.

Robust energy-based least squares twin support vector machines

Twin support vector machine (TSVM), least squares TSVM (LSTSVM) and energy-based LSTSVM (ELS-TSVM) satisfy only empirical risk minimization principle. Moreover, the matrices in their formulations are always positive semi-definite. To overcome these problems, we propose in this paper a robust energy-based least squares twin support vector machine algorithm, called RELS-TSVM for short. Unlike TSVM, LSTSVM and ELS-TSVM, our RELS-TSVM maximizes the margin with a positive definite matrix formulation and implements the structural risk minimization principle which embodies the marrow of statistical learning theory. Furthermore, RELS-TSVM utilizes energy parameters to reduce the effect of noise and outliers. Experimental results on several synthetic and real-world benchmark datasets show that RELS-TSVM not only yields better classification performance but also has a lower training time compared to ELS-TSVM, LSPTSVM, LSTSVM, TBSVM and TSVM.

Enhancing the linearity characteristics of photoelectric displacement sensor based on extreme learning machine method

Photoelectric displacement sensors rarely possess a perfectly linear transfer characteristic, but always have some degree of non-linearity over their range of operation. If the sensor output is nonlinear, it will produce a whole assortment of problems. This paper presents a method to compensate the nonlinearity of the photoelectric displacement sensor based on the extreme learning machine (ELM) method which significantly reduces the amount of time needed to train a neural network with the output voltage of the optical displacement sensor and the measured input displacement to eliminate the nonlinear errors in the training process. The use of this proposed method was demonstrated through computer simulation with the experimental data of the sensor. The results revealed that the proposed method compensated the presence of nonlinearity in the sensor with very low training time, lowest mean squared error (MSE) value, and better linearity. This research work involved less computational complexity, and it behaved a good performance for nonlinearity compensation for the photoelectric displacement sensor and has a good application prospect.

Comparison of classification methods on EEG signals based on wavelet packet decomposition

EEG signals play an important role in both the diagnosis of neurological diseases and understanding the psychophysiological processes. Classification of EEG signals includes feature extraction and feature classification. This paper uses approximate entropy and sample entropy based on wavelet package decomposition as the feature exaction methods and employs support vector machine and extreme learning machine as the classifiers. Experiments are performed in epileptic EEG data and five mental tasks, respectively. Experimental results show that the combination strategy of sample entropy and extreme learning machine has shown great performance, which obtains good classification accuracy and low training time.

An Evolutionary Optimized Algorithm Approach to Compensate the Non-linearity in Linear Variable Displacement Transducer Characteristics

Linearization of transducer characteristic plays a vital role in electronic instrumentation because all transducers have outputs nonlinearly related to the physical variables they sense. If the transducer output is nonlinear, it will produce a whole assortment of problems. Transducers rarely possess a perfectly linear transfer characteristic, but always have some degree of non-linearity over their range of operation. Attempts have been made by many researchers to increase the range of linearity of transducers. This paper presents a method to compensate nonlinearity of Linear Variable Displacement Transducer (LVDT) based on Extreme Learning Machine (ELM) method, Differential Evolution (DE) algorithm and Artificial Neural Network (ANN) trained by Genetic Algorithm (GA). Because of the mechanism structure, LVDT often exhibit inherent nonlinear input-output characteristics. The best approximation capability of optimized ANN technique is beneficial to this. The use of this proposed method is demonstrated through computer simulation with the experimental data of two different LVDTs. The results reveal that the proposed method compensated the presence of nonlinearity in the displacement transducer with very low training time, lowest Mean Square Error (MSE) value and better linearity. This research work involves less computational complexity and it behaves a good performance for nonlinearity compensation for LVDT and has good application prospect.

Impact of patient training patterns on peritonitis rates in a large national cohort study.

Ideal training methods that could ensure best peritoneal dialysis (PD) outcome have not been defined in previous reports. The aim of the present study was to evaluate the impact of training characteristics on peritonitis rates in a large Brazilian cohort. Incident patients with valid data on training recruited in the Brazilian Peritoneal Dialysis Multicenter Study (BRAZPD II) from January 2008 to January 2011 were included. Peritonitis was diagnosed according to International Society for Peritoneal Dialysis guidelines; incidence rate of peritonitis (episodes/patient-months) and time to the first peritonitis were used as end points. Two thousand two hundred and forty-three adult patients were included in the analysis: 59 ± 16 years old, 51.8% female, 64.7% with ≤4 years of education. The median training time was 15 h (IQI 10-20 h). Patients were followed for a median of 11.2 months (range 3-36.5). The overall peritonitis rate was 0.29 per year at risk (1 episode/41 patient-months). The mean number of hours of training per day was 1.8 ± 2.4. Less than 1 h of training/day was associated with higher incidence rate when compared with the intervals of 1-2 h/day (P = 0.03) and >2 h/day (P = 0.02). Patients who received a cumulative training of >15 h had significantly lower incidence of peritonitis compared with <15 h (0.26 per year at risk versus 0.32 per year at risk, P = 0.01). The presence of a caregiver and the number of people trained were not significantly associated with peritonitis incidence rate. Training in the immediate 10 days after implantation of the catheter was associated with the highest peritonitis rate (0.32 per year), compared with training prior to catheter implantation (0.28 per year) or >10 days after implantation (0.23 per year). More experienced centers had a lower risk for the first peritonitis (P = 0.003). This is the first study to analyze the association between training characteristics and outcomes in a large cohort of PD patients. Low training time (particularly <15 h), smaller center size and the timing of training in relation to catheter implantation were associated with a higher incidence of peritonitis. These results support the recommendation of a minimum amount of training hours to reduce peritonitis incidence regardless of the number of hours trained per day.

An efficient genetic selection of the presentation order in simplified fuzzy ARTMAP patterns

Simplified fuzzy ARTMAP (SFAM) is used in numerous classification problems due to its high discriminant power and low training time. However, the performance of SFAM is affected by the presentation order of the training patterns. The genetic algorithm (GA) can be considered as a solution to the problem because the selection of the training pattern order is a complicated combinatorial problem in a large search space. In this paper, a new genetic ordering method for SFAM is proposed to improve the performance of the algorithm. Special genetic operators are employed in the genetic evolution. Compared to the conventional methods, the proposed SFAM demonstrates better classification performance since it can efficiently deliver the desirable properties of parents to their offspring. To demonstrate the performance of the proposed method, we perform experiments on various databases from the UCI repository.

Robust support vector machine-trained fuzzy system

Because the SVM (support vector machine) classifies data with the widest symmetric margin to decrease the probability of the test error, modern fuzzy systems use SVM to tune the parameters of fuzzy if–then rules. But, solving the SVM model is time-consuming. To overcome this disadvantage, we propose a rapid method to solve the robust SVM model and use it to tune the parameters of fuzzy if–then rules. The robust SVM is an extension of SVM for interval-valued data classification.We compare our proposed method with SVM, robust SVM, ISVM-FC (incremental support vector machine-trained fuzzy classifier), BSVM-FC (batch support vector machine-trained fuzzy classifier), SOTFN-SV (a self-organizing TS-type fuzzy network with support vector learning) and SCLSE (a TS-type fuzzy system with subtractive clustering for antecedent parameter tuning and LSE for consequent parameter tuning) by using some real datasets. According to experimental results, the use of proposed approach leads to very low training and testing time with good misclassification rate.

FUZZY EXTREME LEARNING MACHINE FOR A CLASS OF FUZZY INFERENCE SYSTEMS

Recently an Online Sequential Fuzzy Extreme Learning (OS-Fuzzy-ELM) algorithm has been developed by Rong et al. for the RBF-like fuzzy neural systems where a fuzzy inference system is equivalent to a RBF network under some conditions. In the paper the learning ability of the batch version of OS-Fuzzy-ELM, called as Fuzzy-ELM is further evaluated to train a class of fuzzy inference systems which can not be represented by the RBF networks. The equivalence between the output of the fuzzy system and that of a generalized Single-Hidden Layer Feedforward Network as presented in Huang et al. is shown first, which is then used to prove the validity of the Fuzzy-ELM algorithm. In Fuzzy-ELM, the parameters of the fuzzy membership functions are randomly assigned and then the corresponding consequent parameters are determined analytically. Besides an input variable selection method based on the correlation measure is proposed to select the relevant inputs as the inputs of the fuzzy system. This can avoid the exponential increase of number of fuzzy rules with the increase of dimension of input variables while maintaining the testing performance and reducing the computation burden. Performance comparison of Fuzzy-ELM with other existing algorithms is presented using some real-world regression benchmark problems. The results show that the proposed Fuzzy-ELM produces similar or better accuracies with a significantly lower training time.

An application to transient current signal based induction motor fault diagnosis of Fourier–Bessel expansion and simplified fuzzy ARTMAP

The start-up transient signals have been widely used for fault diagnosis of induction motor because they can reveal early defects in the development process, which are not easily detected with the signals in the steady state operation. However, transient signals are non-linear and contain multi components which need a suitable technique to process and identify the fault pattern. In this paper, the fault diagnosis problem of induction motor is conducted by a data driven framework where the Fourier–Bessel (FB) expansion is used as a tool to decompose transient current signal into series of single components. For each component, the statistical features in the time and the frequency domains are extracted to represent the characteristics of motor condition. The high dimensionality of the feature set is solved by generalized discriminant analysis (GDA) implementation to decrease the computational complexity of classification. In the meantime, with the aid of GDA, the separation of the feature clusters is increased, which enables the more classification accuracy to be achieved. Finally, the reduced dimensional features are used for classifier to perform the fault diagnosis results. The classifier used in this framework is the simplified fuzzy ARTMAP (SFAM) which belongs to a special class of neural networks (NNs) and provides a lower training time in comparison to other traditional NNs. The proposed framework is validated with transient current signals from an induction motor under different conditions including bowed rotor, broken rotor bar, eccentricity, faulty bearing, mass unbalance and phase unbalance. Additionally, this paper provides the comparative performance of (i) SFAM and support vector machine (SVM), (ii) SVM in the framework and SVM combined with wavelet transform in previous studies, (iii) the use of FB decomposition and Hilbert transform decomposition. The results show that the proposed diagnosis framework is capable of significantly improving the classification accuracy.

A sparse $${\varvec{L}}_{2}$$ -regularized support vector machines for efficient natural language learning

Linear kernel support vector machines (SVMs) using either $$L_{1}$$ -norm or $$L_{2}$$ -norm have emerged as an important and wildly used classification algorithm for many applications such as text chunking, part-of-speech tagging, information retrieval, and dependency parsing. $$L_{2}$$ -norm SVMs usually provide slightly better accuracy than $$L_{1}$$ -SVMs in most tasks. However, $$L_{2}$$ -norm SVMs produce too many near-but-nonzero feature weights that are highly time-consuming when computing nonsignificant weights. In this paper, we present a cutting-weight algorithm to guide the optimization process of the $$L_{2}$$ -SVMs toward a sparse solution. Before checking the optimality, our method automatically discards a set of near-but-nonzero feature weight. The final objects can then be achieved when the objective function is met by the remaining features and hypothesis. One characteristic of our cutting-weight algorithm is that it requires no changes in the original learning objects. To verify this concept, we conduct the experiments using three well-known benchmarks, i.e., CoNLL-2000 text chunking, SIGHAN-3 Chinese word segmentation, and Chinese word dependency parsing. Our method achieves 1–10 times feature parameter reduction rates in comparison with the original $$L_{2}$$ -SVMs, slightly better accuracy with a lower training time cost. In terms of run-time efficiency, our method is reasonably faster than the original $$L_{2}$$ -regularized SVMs. For example, our sparse $$L_{2}$$ -SVMs is 2.55 times faster than the original $$L_{2}$$ -SVMs with the same accuracy.

Application of CMAC Neural Network to Solar Energy Heliostat Field Fault Diagnosis

Solar energy heliostat fields comprise numerous sun tracking platforms. As a result, fault detection is a highly challenging problem. Accordingly, the present study proposes a cerebellar model arithmetic computer (CMAC) neutral network for automatically diagnosing faults within the heliostat field in accordance with the rotational speed, vibration, and temperature characteristics of the individual heliostat transmission systems. As compared with radial basis function (RBF) neural network and back propagation (BP) neural network in the heliostat field fault diagnosis, the experimental results show that the proposed neural network has a low training time, good robustness, and a reliable diagnostic performance. As a result, it provides an ideal solution for fault diagnosis in modern, large-scale heliostat fields.

Comparison of machine learning techniques for target detection

This paper focuses on machine learning techniques for real-time detection. Although many supervised learning techniques have been described in the literature, no technique always performs best. Several comparative studies are available, but have not always been performed carefully, leading to invalid conclusions. Since benchmarking all techniques is a tremendous task, literature has been used to limit the available options, selecting the two most promising techniques (AdaBoost and SVM), out of 11 different Machine Learning techniques. Based on a thorough comparison using 2 datasets and simulating noise in the feature set as well as in the labeling, AdaBoost is concluded to be the best machine learning technique for real-time target detection as its performance is comparable to SVM, its detection time is one or multiple orders of magnitude faster, its inherent feature selection eliminates this as a separate task, while it is more straightforward to use (only three coupled parameters to tune) and has a lower training time.

Efficient k-Winner-Take-All Competitive Learning Hardware Architecture for On-Chip Learning

A novel k-winners-take-all (k-WTA) competitive learning (CL) hardware architecture is presented for on-chip learning in this paper. The architecture is based on an efficient pipeline allowing k-WTA competition processes associated with different training vectors to be performed concurrently. The pipeline architecture employs a novel codeword swapping scheme so that neurons failing the competition for a training vector are immediately available for the competitions for the subsequent training vectors. The architecture is implemented by the field programmable gate array (FPGA). It is used as a hardware accelerator in a system on programmable chip (SOPC) for realtime on-chip learning. Experimental results show that the SOPC has significantly lower training time than that of other k-WTA CL counterparts operating with or without hardware support.

Dense codes at high speeds: varying stimulus properties to improve visual speller performance

This paper investigates the effect of varying different stimulus properties on performance of the visual speller. Each of the different stimulus properties has been tested in previous literature and has a known effect on visual speller performance. This paper investigates whether a combination of these types of stimuli can lead to a greater improvement. It describes an experiment aimed at answering the following questions. (i) Does visual speller performance suffer from high stimulus rates? (ii) Does an increase in stimulus rate lead to a lower training time for an online visual speller? (iii) What aspect of the difference in the event related potential to a flash or a flip stimulus causes the increase in accuracy. (iv) Can an error-correcting (dense) stimulus code overcome the reduction in performance associated with decreasing target-to-target intervals? We found that higher stimulus rates can improve the visual speller performance and can lead to less time required to train the system. We also found that a proper stimulus code can overcome the stronger response to rows and columns, but cannot greatly improve speller performance.

KFD-Based Multiclass Synthetical Discriminant Analysis for Radar HRRP Recognition

A new extended application of kernel Fisher discriminant (KFD), i.e., KFD-based multiclass synthetical discriminant analysis (KFD-MSDA), is proposed for multi-target discrimination by using the traditional KFD “one-against-one”, and for multi-target classification under the proposed minimum synthetical Euclidian distance (MSED) rule. Theoretical analysis and experimental results on the measured and simulated radar HRRP databases indicate that, compared with several classical kernel-based methods, KFD-MSDA not only performs better with more stable recognition, but also keeps lower training time consuming and competitive recognition speed.

Syntactic Processing Using the Generalized Perceptron and Beam Search

We study a range of syntactic processing tasks using a general statistical framework that consists of a global linear model, trained by the generalized perceptron together with a generic beam-search decoder. We apply the framework to word segmentation, joint segmentation and POS-tagging, dependency parsing, and phrase-structure parsing. Both components of the framework are conceptually and computationally very simple. The beam-search decoder only requires the syntactic processing task to be broken into a sequence of decisions, such that, at each stage in the process, the decoder is able to consider the top-n candidates and generate all possibilities for the next stage. Once the decoder has been defined, it is applied to the training data, using trivial updates according to the generalized perceptron to induce a model. This simple framework performs surprisingly well, giving accuracy results competitive with the state-of-the-art on all the tasks we consider. The computational simplicity of the decoder and training algorithm leads to significantly higher test speeds and lower training times than their main alternatives, including log-linear and large-margin training algorithms and dynamic-programming for decoding. Moreover, the framework offers the freedom to define arbitrary features which can make alternative training and decoding algorithms prohibitively slow. We discuss how the general framework is applied to each of the problems studied in this article, making comparisons with alternative learning and decoding algorithms. We also show how the comparability of candidates considered by the beam is an important factor in the performance. We argue that the conceptual and computational simplicity of the framework, together with its language-independent nature, make it a competitive choice for a range of syntactic processing tasks and one that should be considered for comparison by developers of alternative approaches.

Using genetic algorithm to select the presentation order of training patterns that improves simplified fuzzy ARTMAP classification performance

The presentation order of training patterns to a simplified fuzzy ARTMAP (SFAM) neural network affects the classification performance. The common method to solve this problem is to use several simulations with training patterns presented in random order, where voting strategy is used to compute the final performance. Recently, an ordering method based on min–max clustering was introduced to select the presentation order of training patterns based on a single simulation. In this paper, another single simulation method based on genetic algorithm is proposed to obtain the presentation order of training patterns for improving the performance of SFAM. The proposed method is applied to a 40-class individual classification problem using visual evoked potential signals and three other datasets from UCI repository. The proposed method has the advantages of improved classification performance, smaller network size and lower training time compared to the random ordering and min–max methods. When compared to the random ordering method, the new ordering scheme has the additional advantage of requiring only a single simulation. As the proposed method is general, it can also be applied to a fuzzy ARTMAP neural network when it is used as a classifier.