Many process variables are included in the wastewater treatment process. The realization of real-time detection of as many process variables as possible is of great significance for improve the quality of wastewater treatment. Due to cost constraints, it is difficult to achieve real-time monitoring of some important process variables. Soft measurement scheme, which modeling the relationship between easy-to-measure variables and hard-to-measure variables to achieve the best estimate of the latter is a common solution. This paper proposes a key variables soft measurement scheme for wastewater treatment process based on hierarchical extreme learning machine (HELM). The proposed scheme takes some of the known variables that are easy to measure as inputs, and then implement a best estimate of unknown variables that are difficult to measure. At the same time, the selective moving window strategy (SMW) is used to update the training datasets. Experiments show that the proposed scheme has excellent performance on many key indicators.
In recent years, event-based social network (EBSN) platforms have increasingly entered people’s daily life and become more and more popular. In EBSNs, event recommendation is a typical problem which recommends interested events to users. Different from traditional social networks, both online and off-line factors play an important role in EBSNs. However, the existing methods do not make full use of the online and off-line information, which may lead to a low accuracy, and they are also not efficient enough. In this paper, we propose a novel event recommendation model to solve the above shortcomings. At first, a feature extraction phase is constructed to make full use of the EBSN information, including spatial feature, temporal feature, semantic feature, social feature and historical feature. And then, we transform the recommendation problem to a classification problem and ELM is extended as the classifier in the model. Extensive experiments are conducted on real EBSN datasets. The experimental results demonstrate that our approach is efficient and has a better performance than the existing methods.
Bidirectional extreme learning machine (B-ELM) divides the learning process into two parts: At odd learning step, the parameters of the new hidden node are generated randomly, while at even learning step, the parameters of the new hidden node are obtained analytically from the parameters of the former node. However, some of the odd-hidden nodes play a minor role, which will have a negative impact on the even-hidden nodes, and result in a sharp rise in the network complexity. To avoid this issue, we propose a random orthogonal projection based enhanced bidirectional extreme learning machine algorithm (OEB-ELM). In OEB-ELM, several orthogonal candidate nodes are generated randomly at each odd learning step, only the node with the largest residual error reduction will be added to the existing network. Experiments on six real datasets have shown that the OEB-ELM has better generalization performance and stability than B-ELM, EB-ELM, and EI-ELM algorithms.
The big data trend has inspired feature-driven learning tasks, which cannot be handled by conventional machine learning models. Unstructured data produces very large binary matrices with millions of columns when converted to vector form. However, such data is often sparse, and hence can be manageable through the use of sparse random projections. This work studies efficient non-iterative and iterative methods suitable for such data, evaluating the results on two representative machine learning tasks with millions of samples and features. An efficient Jaccard kernel is introduced as an alternative to the sparse random projection. Findings indicate that non-iterative methods can find larger, more accurate models than iterative methods in different application scenarios.
With the increasing popularity of location-based social networks (LBSNs), an increasing number of people are sharing their locations with friends through check-in activities. Point-of-interest (POI) recommendation, in which new places are suggested to users, is one of the most important tasks in LBSNs. POI group recommendation aims to suggest the most agreeable places for a group of users. However, the most existing studies, based on collaborative filtering, suffered from various issues, such as: (1) data sparseness, (2) cold start, and (3) scalability. Moreover, many existing schemes are limited in functionality. To address the aforementioned issues, we proposes a POI group recommendation model from a new perspective. The model combines machine learning knowledge and considers POI group recommendation as a classification problem. Extreme learning machine (ELM) is used to train the features. ELM has fast learning speed and ensures the recommendation efficiency. Finally, a series of experiments verify the performance of the model with ELM.
In order to pursue efficiency improvement for vehicles, embedding the information of connected vehicle and infrastructure to the powertrain system control is taken as potential ways. This paper proposes an integrated control framework to deal with the energy consumption optimization problem for a hybrid electric vehicle (HEV). Taking the vehicle motion dynamics as constraint, a receding horizon control (RHC) scheme is developed to get a short-term optimal solution of power distribution for the HEV. To further improve the vehicle performance, an extreme learning machine (ELM) model is developed with the information of vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) to predict the real-time demand torque variation. The performance of the control system is finally demonstrated at a traffic-in-the-loop powertrain simulation system (TILPSS) platform.
In modern manufacturing industry featured with automation and flexibility, the intelligent machine tools management is essential for the workshop. In this work, we proposed a novel machine tools recognition system for classifying 3D models. A common and standard 3D tool database is constructed. The hybrid networks of Convolutional Neural Networks (CNNs) and Extreme Learning Machine (ELM) are developed for multiple view based 3D shape recognition. This framework utilizes the composited advantages of deep CNN architecture with the robust ELM auto-encoder feature representation, as well as the fast ELM classifier. The experimental results shows that it outperforms other methods which are using the manually specified 3D feature descriptors.
Extreme Learning Machine (ELM) is a popular method in machine learning with extremely few parameters, fast learning speed and model efficiency. While a significant drawback is that ELM is restricted by its single-layer structure and prized analytic solution. If simply stacking more layers, analytic solution of ELM will be intractable. Then gradient-based optimization method is preferred and that results into normal neural networks. Recently a multi-layer ELM (ML-ELM) is proposed to learn compact feature with a series of ELM auto-encoders, which attempts to extend ELM to a deeper network without sacrificing elegant solution. Compared with ML-ELM and following hierarchical ELM, we introduce a sparse Bayesian learning method to imply a stronger sparse regularization and prune network structure. Experiments on classification verify the efficiency of our proposed new multi-layer ELM for unsupervised feature learning.
Complex power quality disturbances (CPQDs) classification can be regarded as a typical application of multi-label (ML) learning. In this study, we propose a new recognition method for CPQDs based on S-transform (ST) and a hybrid kernel function-based extreme learning machine (ELM) for ML learning (HKEML). The signal processing techniques S-transform is utilized to extract the distinctive features of the CPQDs. A novel ML classifier called HKEML is constructed by combining hybrid kernel function-based multiclass ELM and a thresholding learning method-based kernel ELM. Finally, a test study was conducted using Matlab synthetic signals and real signals sampled from a three-phase standard source under different noise conditions. Compared with several recent state-of-the-art ML learning algorithms, HKEML achieved better classification performance but with greatly superior computational speed.
Extreme Learning Machine (ELM) has drawn more and more attention in the machine learning fields because of its fast training speed and good generalization ability. In this paper, we employ ELM algorithm to deal with multi-label classification problems. The essence of the proposed algorithm is to convert the multi-label classification problem into some single-label classifications, and fully considers the relationship among different labels. The hidden nodes of ELM algorithm for one label classification come from three parts. Some of them are initialized nodes after PCA dimensionality reduction. And we design a backup pool to select appropriate hidden nodes. In addition, the nodes of the previous label classification would flow into the classification of the current label. In the simulation part, three famous databases demonstrate the satisfied classification accuracy of the proposed method.
As an important branch of neural networks, extreme learning machine with single-hiddenlayer feedforward have been a effective tool for regression and classification applications. However, it is difficult for ELMs to strike a balance between testing accuracy and generalization due to the random input weights and hidden biases. In this paper, a novel multi-objective optimization method of ELM based on swarm intelligence behavior is proposed to obtain good generalization ability and high testing accuracy simultaneously. The multi-objective optimization algorithm is used to select optimal input weights by minimizing this testing error and the norm of output weight. In order to improve optimal performance, an information learning method is introduced to multi-objective artificial bee colony algorithm. Experiments on four UCI data sets are conducted, and original ELM, ELM with nondominated sorting genetic algorithm and the proposed algorithm are compared. The results show that the proposed algorithm can generally obtain better generalization performance and higher accuracy with more compact network than original ELM and ELM with nondominated sorting genetic algorithm simultaneously.
The usual computer aided diagnosis approaches of Alzheimer’s disease patients based on fMRI often require a lot of manual intervention. By contrast, H-ELM needs only less manual intervention and can extract features by a multi-layer feature representation framework. Therefore, an AD CADx model based on H-ELM is proposed. First, the common spatial pattern is used to extract information from the BOLD signals, and then the features are encoded and trained by H-ELM. H-ELM is used to realize the expression of deep feature of the brain, so as to further improve the diagnostic accuracy. Finally, experimental evaluation proved the effectiveness of the proposed algorithms.
Representational learning using Extreme Learning Machine (ELM) theory has aroused lots of interest. Motivated by recent advance in ELM, this paper presents a novel Extreme Latent Representation (ELR) learning model to seamlessly connect original perception data and the corresponding high-level semantics. Specifically, ELM and ELM based auto-encoder (ELM-AE) are formulated in a unified learning model with both classification and reconstructive ability of the representation considered. ELR inherits the merits of ELM and ELM-AE, and discriminative and compact representation can be learnt with data information well preserved. Furthermore, an efficient algorithm based on alternating direction method of multipliers (ADMM) is developed to solve the resulting ELR model. The performance of ELR is verified on two visual classification tasks, and encouraging results have are achieved.
The prosperity of Location-Based Social Networks (LBSNs) facilitates a promising focus on personalized POI recommendation. The check-in activity is a multiple criteria decision making process. In this paper, we devise a novel personalized POI recommendation system called Multiple Criteria Ensemble Recommendation System (MCERS) to integrate the effect of geographical influence, temporal influence and categorical influence in a unified way to model dynamic user preferences for context-aware query. Specifically, we propose an ELM-based ensemble strategy to provide a more sophisticated framework for integrating different criteria to improve the quality of POI recommendations. Extensive experiments on three real-world datasets show that MCERS achieves better performance than existing state-of-the-art methods.
Rising stars are junior individuals in social networks who will have high impacts with time accumulation. In this paper, we study the problem of rising star evaluation in geo-social networks. Specifically, given a topic keyword, we aim at evaluating the latent influence of users to find rising star, which refer to expert that may have few activities and little impact currently in the underlying geo-social network, but he or she will become influential experts in the future. Most of the current research studies focus on experts finding rather than rising stars. Rising stars can bring great opportunities. We investigate a novel processing framework based on extreme learning machine (ELM) called FS-ELM to efficiently evaluate future stars. Our experimental studies conducted on real-world datasets demonstrate that our method is efficient in predicting rising star with potential impact.
As a popular classification algorithm for machine learning, Extreme Learning Machine (ELM) has been widely used. However, its performance on various hardware devices is unclear. According to the baseline implementation of single core ELM, we find that the main time cost of ELM is matrix multiplication. Then, this paper designs various optimized hardware algorithms for several computing devices (Multi-Core, GPU, and FPGA). According to the experiment of each platform, we can see that the speedup ratio of the new hardware platform to ELM is 4~100+, we open our source code and strongly recommend that the later researchers design the application of ELM algorithm based on appropriate hardware platform.
Precise daily rainfall forecasting play a very significant role in modern society that it can not only help for planning of people’s day-to-day activities, agriculture and business, but also assist water resource management in the region to warn or alleviate the effect of drought or flood disaster. However, various inherently complex meteorological factors and dynamic behavior influence the rainfall, with result that it is very difficult to accurately forecast daily rainfall. This study presents a soft computing modeling method based on Extreme Learning Machine (ELM) and Gene Expression Programming (GEP) to enhance the forecast performance. The proposed mode is compared with other five rainfall forecasting models to assess its performance for rainfall forecasting by Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Experimental results show that the proposed method outperforms other models in terms of accuracy.
Text categorization, or text classification, is one of key tasks for representing the semantic information of documents. Traditional deep leaning models for text categorization are generally time-consuming with large-sized datasets due to slow convergence rate. In this paper, we propose a character-level model for short text classification with a combination of convolutional neural network (CNN), gated recurrent unit (GRU) and highway network (HN), which can capture both the global and the local textual semantics while having a tractable computational complexity. In addition, error minimization extreme learning machine (EM-ELM) is incorporated into the proposed model to improve the classification accuracy further. Extensive experiments show that our approach achieves the state-of-the-art performance when the hybrid model based on EM-ELM is trained using large-sized datasets.
Learning from the absent multiple sources data is very challenging, especially when lacking label information. Although existing multiple kernel learning methods with kernel imputing achieved remarkable performance, they may fail in practical issues due to hardly tracking the data with huge missing samples or easily fall into a local optimum. Thus, we propose an unsupervised absent multiple kernel extreme learning machine to effectively deal with absent data. It firstly extracts information from absent multiple sources data via multiple kernels to structure a new data space named K-space. Then, the optimal multiple kernel combination coefficients are learned in the K-space. Finally, we establish a seamlessly integrated object function to complete the absent kernel matrices and acquire the clustering information simultaneously by a well-designed optimization strategy within finite steps. As evidenced by comprehensive experiments, the proposed method provides significantly better clustering performance.
