# Deep Network Based on Stacked Orthogonal Convex Incremental ELM Autoencoders.

1. IntroductionExtreme learning machine (ELM) proposed by Huang et al. [1,2] is a specific type of single-hidden layer feedforward network (SLFN) with randomly generated additive or RBF hidden nodes and hidden node parameters, which has recently been extensively studied by many researchers in various areas of scientific research and engineering due to the excellent approximation capability. Wang et al. presented ASLGEMELM algorithm, which provides some useful guidelines for improving the generalization ability of SLFNs trained with ELM [3]. Alongside probing deeply into the research of theory and its application, ELM has become one of the leading trends for fast learning [4-7]. Recently, Huang et al. [8] have proposed an algorithm called incremental extreme learning machine (I-ELM) which randomly adds nodes to the hidden layer one by one and freezes the output weights of the existing hidden nodes when a new hidden node is added [9-12]. Then, Huang et al. [13] also showed its universal approximation capability for the case of fully complex hidden nodes. I-ELM is fully automatically implemented and in theory no intervention is required for the learning process from users. But there still exist some issues to be tackled [14]:

(1) The redundant nodes can be generated in I-ELM, which have a minor effect on the outputs of the network. Moreover, the existence of redundant nodes can eventually increase the complexity of the network.

(2) The convergence rate of I-ELM is slower than ELM, and the number of hidden nodes in I-ELM is sometimes larger than the dimension of samples for the training.

In this paper, we propose a method called orthogonal convex extreme learning machine (OCI-ELM) to further settle the aforementioned problems of I-ELM. With the rigorous proofs in theory, we can obtain the least-squares solution of H[beta] = T and faster convergence rate by adopting the Gram-Schmidt orthogonalization method incorporated into CI-ELM [15]. The simulations on real-world datasets show that the proposed OCI-ELM algorithm can achieve faster convergence rates, more compact neural network, and better generalization performance than both I-ELM and the improved I-ELM algorithms while keeping the simplicity and efficiency of ELM.

Recently, deep learning has attracted many research interests with its remarkable success in many applications [16-18]. Deep learning is an artificial neural network learning algorithm which has multilayer perceptrons. Deep learning has achieved an approximation of complex functions and alleviated the optimization difficulty associated with the deep models [19-21]. Motivated by the remarkable success of deep learning [22, 23], we propose a new stacked architecture to solve large and complex data problems using OCI-ELM autoencoder as the training algorithm in each layer, which incorporates the excellent performance of OCI-ELM with the ability of complex function approximation derived from deep architectures. We implemented OCI-ELM autoencoder in each iteration of deep orthogonal convex incremental extreme machine (DOC-IELM) to reconstruct the input data and estimate the errors of the prediction functions with the scheme of layer-by-layer architectures. Both the supervised and the unsupervised data all can be the pertaining input of the new proposed deep network. Moreover, the OCI-ELM autoencoder-based deep network (DOC-IELM-AEs) cansuffice to achieve the efficiency improvement for generalization performance.

To show the effectiveness of DOC-IELM-AEs, we apply it to both the ordinary real-world datasets with UCI datasets and large datasets with MNIST, OCR Letters, NORB, and USPS datasets. The simulations show that the proposed deep model possesses better accuracy of testing and more compact network architecture than the aforementioned improved IELM and other deep models without incurring the out-of-memory problem.

This paper is organized as follows. Section 2 reviews the preliminary knowledge of incremental extreme learning machine (I-ELM). Section 3 describes OCI-ELM algorithm, the proposed model which adopts the Gram-Schmidt orthogonalization method into convex I-ELM (CI-ELM). Section 4 makes a comparison between OCI-ELM and other algorithms. Section 5 presents the details of DOC-IELM-AEs algorithm and compares the performance with deep architecture models. Section 6 applies the DOC-IELM-AEs algorithm into elongation prediction of strips. Finally, Section 7 concludes this paper.

2. Related Works

In this section, the main concepts and theory of the I-ELM [8] algorithm are shortly reviewed. For the sake of generality, we assume that the network has only one linear output node, and all the analysis can be easily extended into multinonlinear output nodes cases. Consider a training dataset N = {([x.sub.i], [t.sub.i]) | [x.sub.i] [member of] [R.sup.n], [t.sub.i] [member of] [R.sup.m], 1 {less than or equal to] i [less than or equal to] N}; the SLFN with L additive hidden nodes and activation function [sigma](x) can be represented by

[f.sub.n](x) = [L.summation over (i=1)] [[beta].sub.i][sigma]([a.sub.i] * [x.sub.j] + [b.sub.j]), j = 1,2, ..., N, (1)

where [a.sub.i] = [[[a.sub.i1], [a.sub.i2], ..., [a.sub.in]].sup.T] is the weight vector connecting the input layer to the ith hidden node, [[beta].sub.i] = [[[[beta].sub.i1], [[beta].sub.i2], ..., [[beta].sub.in]].sup.T] is the weight connecting the ith hidden node to the output node, [b.sub.i] is the threshold of the ith hidden node, and [sigma] is the hidden node activation function.

The I-ELM proposed by Huang et al. is different from the conventional ELM algorithm; I-ELM is an automatic algorithm which can randomly add hidden nodes to the network one by one and freeze all the weights of the existing hidden nodes when a new hidden node is added, until the expected learning accuracy is obtained or the maximum number of hidden nodes is reached. Thus, I-ELM algorithm can be summarized in Algorithm 1.

Algorithm 1 (incremental extreme learning machine (I-ELM)). Given a training dataset N = {([x.sub.i], [t.sub.i]) | [x.sub.i] [member of] [R.sup.n], [t.sub.i] member of] [R.sup.m], 1 [less than or equal to] i [less than or equal to] N}, activation function g(x), number of hidden nodes L, expected learning accuracy [epsilon], and maximum number of hidden nodes [L.sub.max], one has the following.

Step 1 (initialization). Let L = 0 and residual error E = t, where t = [[[t.sub.1], [t.sub.2], ..., [t.sub.N]].sup.T].

Step 2 (learning step). While L < [L.sub.max], [parallel]E[parallel] > [epsilon],

(a) increase the number of hidden nodes by one;

(b) assign random input weight w and bias b for hidden nodes L;

(c) calculate the residual error after adding the new hidden node;

(d) calculate the output weight [beta] for the new hidden nodes: [beta] = (E, [H.sub.L])/[parallel][H.sub.L][[parallel].sup.2];

(e) calculate the residual error: E = E - [beta] * [H.sub.L];

Endwhile.

3. The Proposed Orthogonal Convex Incremental Extreme Learning Machine (OCI-ELM)

The motivation for the work in this section comes from the important properties of basic ELM as follows:

(1) The special solution [??] = [H.sup.[dagger]]T is one of the least-squares solutions of a general linear system H[beta] = T, meaning that the smallest training error can be reached by this special solution: [parallel]H[??]-T[parallel] = [parallel]H[H.sup.[dagger]]T-T[parallel] = [min.sub.[beta]][parallel]H[beta] - T[parallel].

(2) The smallest norm of weights: the special solution [??] = [H.sub.[dagger]]T has the smallest norm among all the least-squares solutions of H[beta] = T:

[mathematical expression not reproducible]. (2)

(3) The minimum norm least-squares solution of H[beta] = T is unique, which is [??] = [H.sup.[dagger]]T.

In this section, we propose an improved I-ELM algorithm (OCI-ELM) based on Gram-Schmidt orthogonalization method combined with Barron's convex optimization learning method and prove the OCI-ELM algorithm in theory which can obtain the least-squares solution of H[beta] = T. Meanwhile, OCI-ELM can achieve a more compact network architecture, faster convergence rate, and better generalization performance than other improved I-ELM algorithms while retaining the I-ELM's simplicity and efficiency.

Theorem 2. Gram-Schmidt orthogonalization process converts linearly independent vectors into orthogonal vectors [24]. Given a linearly independent vector set [[[alpha].sub.1], [[alpha].sub.2], ..., [[alpha].sub.n]] in the inner product space V [subset or equal to] [R.sup.n], the vector set [[beta].sub.1], ..., [[beta].sub.2], ..., [[beta].sub.n] for Gram-Schmidt orthogonalization process is as follows [25]:

[mathematical expression not reproducible], (3)

where [[alpha].sub.1], [[alpha].sub.2], ..., [[alpha].sub.n] is the set of standardized vectors and form an orthogonal set with the same linear span. For each index k = 1,2, ..., n, span([[beta].sub.1], [[beta].sub.2], ..., [[beta].sub.n]) = span([[alpha].sub.1], [[alpha].sub.2], ..., [[alpha].sub.n]).

Theorem 3. Givenanorthogonal vectorset [[v.sub.1], [v.sub.2], ..., [v.sub.k-1]] in the inner product space V [subset or equal to] [R.sup.n], if vector [g.sub.k] can be expressed as a linear representation of [v.sub.1], [v.sub.2], ..., [v.sub.k-1], one has

[mathematical expression not reproducible]. (4)

Proof. Given the vector set [v.sub.1], [v.sub.2], ..., [v.sub.k-1] and vector [g.sub.k], suppose there exist scalars [c.sub.1], [c.sub.2], ..., [c.sub.k-1]; then the linear combination of those vectors with those scalars as coefficients is

[g.sub.k] = [c.sub.1][v.sub.1] + [c.sub.2][v.sub.2] + *** + [c.sub.k-1][v.sub.k-1]. (5)

Substituting (4) into (5), we have [v.sub.k] = 0:

[[beta].sub.n] = ([e.sub.n-1], [g.sub.n] - [f.sub.n-1])/ [parallel][g.sub.n] - [f.sub.n-1][[parallel].sup.2]. (6)

CI-ELM was originally proposed by Huang and Chen [15], which incorporates Barron's convex optimization learning method into I-ELM. By recalculating the output weights of the existing hidden nodes randomly generated after a new node is added, the CI-ELM can obtain better performance than I-ELM. Incorporated with Gram-Schmidt orthogonalization and Barron's convex optimization learning method, the process of OCI-ELM algorithm can be described in Algorithm 4.

Algorithm 4 (orthogonal convex incremental extreme learning machine (OCI-ELM)). Given a training dataset N = {([x.sub.i], [t.sub.i]) | [x.sub.i] [member of] [R.sup.n], [t.sub.i] [member of] [R.sup.m], i = 1,2, ..., N}, where [x.sub.i] = [[[x.sub.i1], [x.sub.i2], ..., [x.sub.in]].sup.T] and [t.sub.i] = [[t.sub.i1], [t.sub.in], ..., [t.sub.im]], and given activation function [sigma](x), maximum number of iterations [K.sub.max], and expected learning accuracy [epsilon], one has the following.

Step 1 (initialization). Let the number of initial hidden nodes L = 0, the number of iterations K = 0, and residual error E = t, where t = [[[t.sub.1], [t.sub.2], ..., [t.sub.N]].sup.T].

Step 2. This step consists of two steps as follows.

Orthogonalization Step. In this step, the following is carried out:

(a) Increase the number of hidden nodes L and K by one, respectively: L = L + 1 and K = K + 1.

(b) Randomly assign hidden node parameters ([a.sub.L], [b.sub.L]) for new hidden node L and calculate the output [g.sub.k],

[mathematical expression not reproducible], (7)

and the hidden layer output matrix V,

[mathematical expression not reproducible]. (8)

Learning Step. While [L.sub.1] < [L.sub.max], [parallel]E[parallel] > [epsilon],

(c) calculate the output weight [[beta].sub.k] for the newly added hidden node:

[[beta].sub.k] = E * [[E - (F - [V.sub.k])].sup.T]/ [E - (F - [V.sub.k])] * [[E - (F - [V.sub.k])].sup.T]; (9)

(d) recalculate the output weight vectors of all existing hidden nodes if L > 1:

[[beta].sub.l] = (1 - [[beta].sub.k])[[beta].sub.l], I = 1, 2, ..., L - 1; (10)

(e) calculate the residual error after adding the new hidden node L:

E = (1 - [[beta].sub.k])E + [[beta].sub.k] (F - [V.sub.k]); (11)

Endwhile.

The rigorous proof on the conclusion is detailedly discussed where OCI-ELM can obtain the least-squares solution of H[beta] = T.

Theorem 5. Given a training dataset N = {([x.sub.i], [t.sub.i]) | [x.sub.i] [member of] [R.sup.n], [t.sub.i] [member of] [R.sup.m], i = 1, 2, ..., N] and number of hidden nodes L, where [x.sub.i] = [[[x.sub.1], [x.sub.2], ..., [x.sub.in]].sup.T] and [t.sub.i] = [[[t.sub.1], [t.sub.2], ..., [t.sub.im]].sup.T], the hidden layer output matrix is [G.sub.L] = [[g.sub.1], [g.sub.2], ..., [g.sub.L]] [member of] [R.sup.nxL], and the matrix of the output weights from the hidden nodes to the output nodes is [[beta].sub.L] = [[[[beta].sub.1], [[beta].sub.2], ..., [[beta].sub.L]].sup.T] [member of] [R.sup.L]. Let [e.sub.n] = [e.sub.n-1] - [[beta].sub.n]([g.sub.n] - [f.sub.n-1]) denote the residual error function, and [e.sub.0] = T, [[beta].sub.k] = {[[beta].sup.*] | min[parallel][e.sub.k][parallel]} = {[[beta].sup.*] | min[parallel] f - ((1 - [[beta].sub.k]))[f.sub.k-1] + [[beta].sub.k][g.sub.k][parallel]} holds with probability one if [e.sub.L] [perpendicular to] span{[g.sub.1], [g.sub.2], ..., [g.sub.L]} and [[beta].sub.k] = ([e.sub.k-1], [g.sub.k] - [f.sub.k-1])/[parallel][g.sub.k] - [f.sub.k-1][[parallel].sup.2] for all k = 1, 2, ..., L.

Proof. The proof consists of two steps:

(a) Firstly, we prove [e.sub.L] [perpendicular to] span{[g.sub.1], [g.sub.2], ..., [g.sub.L]}.

(b) And then, we further prove [[beta].sub.k] = {[[beta].sup.*] | min[parallel]f - ((1 [[beta].sub.k]))[f.sub.k-1] + [[beta].sub.k][g.sub.k][parallel]}.

(a) According to the condition given above, we have the following:

(1) Here,

([e.sub.1], [g.sub.1]) = {T - [[beta].sub.1][g.sub.1],[g.sub.1]) = (T,[g.sub.1]) - [beta] - 1 {[g.sub.1],[g.sub.1]) = 0. (12)

(2) When the output weight [beta] = {[e.sub.1], [g.sub.2] - [f.sub.1])/[parallel][g.sub.2] - [f.sub.1][[parallel].sup.2], we have

[mathematical expression not reproducible]. (13)

(3) When the output weight [beta] = ([e.sub.1], [g.sub.2] - [f.sub.1])/[parallel][g.sub.2] - [f.sub.1][[parallel].sup.2], we also have

[mathematical expression not reproducible]. (14)

(4) When the output weight [[beta].sub.k] = {[e.sub.k-1], [g.sub.k] - [f.sub.k-1])/[parallel][g.sub.k] [f.sub.k-1][[parallel].sup.2], suppose that, for all k [less than or equal to] l, we have

[mathematical expression not reproducible]. (15)

(5) When the output weight [[beta].sub.k] = {[e.sub.k-1], [g.sub.k] - [f.sub.k-1])/[parallel][g.sub.k] [f.sub.k-1][[parallel].sup.2], suppose that, for all j [less than or equal to] k - 1, we have

[mathematical expression not reproducible]. (16)

So, [e.sub.k] [perpendicular to] span{[g.sub.1], [g.sub.2], ..., [g.sub.k]}; that is, [e.sub.L] [perpendicular to] span{[g.sub.1], [g.sub.2], ..., [g.sub.L]}. Therefore,

[G.sup.T.sub.L] (T - [G.sub.L] (1 - [[beta].sub.L])[[beta].sup.*]) = 0. (17)

(b) According to (17), we have [([[beta].sub.L] - [[beta].sup.*]).sup.T] [G.sup.T.sub.L](T - [G.sub.L](1 [[beta].sub.L])[[beta].sup.*])= 0, where [[beta].sup.*] [member of] [R.sup.Lx1] is arbitrary; then, we have

[mathematical expression not reproducible]. (18)

And [parallel]T - [G.sub.L](1 - [[beta].sub.L])[[beta].sup.*][parallel] = [parallel]T - [G.sub.L](1 - [[beta].sub.L])[[beta].sub.L][parallel] holds only if [[beta].sub.L] = [[beta].sup.*]. Therefore, [[beta].sub.k] = {[[beta].sup.*] | min[parallel]f - ((1 - [[beta].sub.k]))/[f.sub.k-1] + [[beta].sub.k][g.sub.k][parallel]}.

4. Experiments and Analysis

In this section, we tested the generalization performance of the proposed OCI-ELM with other similar learning algorithms on ten UCI real-world datasets, including five regression and five classification problems, as shown in Table 1. The simulations are conducted in MATLAB 2013a environment running on Windows 7 machine with 32 GB of memory and i7-990X (3.46 GHz) processor.

The experimental results between OCI-ELM and some other ELM algorithms on regression and classification problems are given in Tables 2 and 3. In Tables 2 and 3, the best results obtained by the OCI-ELM and the other 4 algorithms are italicized and shown in boldface. In Section 4.1, we compare the generalization performance of OCI-ELM with another six state-of-the-art algorithms on regression problems. In Section 4.2, we compare the generalization performance of OCI-ELM with the same six algorithms on classification problems. All these results in this section are obtained from thirty trials for all cases, and the mean results (mean), root-mean-square errors (RMSE), and standard deviations (Std.) are listed in the corresponding tables, respectively. The seven representative evolutionary algorithms are listed as follows:

(i) Convex incremental extreme learning machine (CI-ELM) [15].

(ii) Parallel chaos search based incremental extreme learning machine (PC-ELM) [26].

(iii) Leave-one-out incremental extreme learning machine (LOO-IELM) [27].

(iv) Sparse Bayesian extreme learning machine (SB-ELM) [28].

(v) Improved incremental regularized extreme learning machine (II-RELM) [11].

(vi) Enhancement incremental regularized extreme learning machine (EIR-ELM) [12].

4.1. Performance Comparison of Regression Problems. In this section, datasets Auto MPG, California Housing, Servo, CCS (Concrete Compressive Strength), and Parkinsons are conducted for the regression problems. Table 2 shows the RMSE of the training and testing with fixed hidden nodes obtained from OCI-ELM and another six algorithms, respectively. Meanwhile, hidden nodes and learning time with the same stop RMSE are also shown in Table 2. For California Housing dataset in the table, the OCI-ELM provides lower training and testing RMSE rate (0.1272 and 0.1263) than CI-ELM (0.1601 and 0.1583), PC-ELM (0.1389 and 0.1377), LOO-IELM (0.1376 and 0.1374), SB-ELM (0.1363 and 0.1369), II-RELM (0.1341 and 0.1339), and EIR-ELM (0.1274 and 0.1268) with fixed nodes (20). For the stop criterion of RMSE 0.12, OCI-ELM also exhibits more compact network architecture with 127.15 nodes and faster speed with 0.9704 s, and the nodes and training speed of other algorithms are, respectively, 330.09 and 1.0051; 199.34 and 0.9810; 217.08 and 0.9766; and 192.33 and 0.9713. Where SB-ELM is the fixed ELM, thus there is a difficulty in finding the accurate stop criterion. Likewise, in the CCS dataset, the hidden node of SB-ELM is an approximate value; meanwhile, OCIELM shows better generalization performance than other algorithms in comparisons. Although, in the cases of Auto MPG, Servo, and Parkinsons, the learning time consumed by OCI-ELM shows that the presented algorithm is not the top spot, the average convergence rate of five regression problems consumed by OCI-ELM is still the fastest. Moreover, the average convergence rate demonstrates that the stability performance of OCI-ELM is better than other algorithms. The proposed algorithm can retain the simplicity and efficiency of incremental ELM and obtain the least-squares solution of H[beta] = T by incorporating the Gram-Schmidt orthogonalization method. The optimal solution obtained from H[beta] = T means that the best hidden node parameter leading to the largest residual error decreasing will be added to the existing network. Therefore, OCI-ELM can efficiently reduce the network complexity and meanwhile enhance the generalization performance of the algorithm.

4.2. Performance Comparison of Classification Problems. In this section, datasets Delta Ailerons, Waveform II, Abalone, Breast Cancer, and Energy Efficiency are conducted for the classification problems. Table 3 shows the comparisons of the classification performance conducted on 5 UCI datasets of classification problems. With the same fixed hidden nodes listed in Table 3, the results of comparisons obtained from OCI-ELM are better than those of the other algorithms. For Waveform II dataset in the table, the OCI-ELM proposed displays better training accuracy and standard deviation of 93.11% and 0.0083 than CI-ELM (84.47% and 0.0182), PC-ELM (89.81% and 0.0104), LOO-IELM (88.93% and 0.0097), SB-ELM (80.69% and 0.0181), II-RELM (90.64% and 0.0112), and EIR-ELM (91.15% and 0.0096), thanks to the better classification ability of OCI-ELM. In addition, the hidden nodes 29.54 and the average time 3.0864 s are also less than others, which can demonstrate that OCI-ELM has more reasonable network structure than CI-ELM, PC-ELM, LOO-IELM, SB-ELM, II-RELM, and EIR-ELM, which efficiently reduces the complexity of the network. Although SB-ELM shows the advantages in training speed as the fixed ELM, OCI-ELM generally produces better performance in comprehensive consideration of accuracy and speed for practical problems which need higher accuracy demand.

In short, OCI-ELM can generally achieve better performance on these regression and classification problems in terms of training (and testing) RMSE for regression and testing accuracy for classification. Moreover, the compactness of network and convergence rate also display the good performance of OCI-ELM algorithm.

5. Deep Network Based on Stacked OCI-ELM Autoencoders (DOC-IELM-AEs)

5.1. OCI-ELM Autoencoder. As an artificial neural network model, the autoencoder is frequently applied in the deep architecture approaches. Autoencoder is a kind of unsupervised neural network, where the input of network is equal to the output. Kasun et al. [29] proposed an autoencoder based on ELM (ELM-AE). According to their ELM-AE theory, the model of ELM-AE is composed of input layer, hidden layer, and output layer. In addition, the weights and biases of the hidden nodes are randomly generated via orthogonalization, and the input data is projected to a different or equal dimension space [30]; the expressions are as follows:

[mathematical expression not reproducible], (19)

where a = [[[a.sub.1], ..., [a.sub.L]].sup.T] are the weights generated orthogonally randomly and b = [[[b.sub.1], ..., [b.sub.L]].sup.T] are the biases generated orthogonally randomly between the input and hidden nodes. There are three calculation approaches to obtain the output weight [beta] of ELM-AE:

(1) For sparse ELM-AE representations, output weights [beta] can be calculated as follows:

[beta] = [(I/C + [H.sup.T]H).sup.-1] [H.sup.T] X. (20)

(2) For compressed ELM-AE representations, output weights [beta] can be calculated as follows:

[beta] = [H.sup.T] [(I/C + [H.sup.T]H).sup.-1] X. (21)

(3) For equal dimension ELM-AE representations, output weights [beta] can be calculated as follows:

[beta] = [H.sup.-1] X. (22)

In this section, we present the OCI-ELM, which is incorporated with Barron's convex optimization learning method and Gram-Schmidt orthogonalization method into I-ELM to achieve the optimal least-squares solution as the training algorithm for an autoencoder instead of conventional autoencoders, which apply backpropagation algorithm (BP) for training to obtain the identity function and normal ELM for training the autoencoder. Because of the adoption of incremental algorithm, there is no need to set the number of hidden nodes according to the experience. With the initialization of the maximum value [L.sub.max] of hidden nodes the, number of hidden nodes can be increased by more than one node each time, until the stop criterions are met; for example, residual error E is equal to the expected learning accuracy e or the number of hidden nodes L achieves [L.sub.max].

As shown in Figure 1, the model structure of OCI-ELM-AE can randomly control the number of the nodes without the computation accuracy. Given a training dataset N = {([x.sub.i], [t.sub.i]) | [x.sub.i] [member of] [R.sup.n], [t.sub.i] [member of] [R.sup.m], i = 1, ..., N}, where [x.sub.i] = [[[x.sub.i1], [x.sub.i2], ..., [x.sub.in]].sup.T] and [t.sub.i] = [[[t.sub.i1], [t.sub.i2], ..., [t.sub.im]].sup.T,] and given activation function [sigma](x) and maximum number of hidden nodes in single layer [L.sub.max], the input data is reconstructed at the output layer through the following function:

[l.summation over (i=1)][[beta].sub.i][sigma] ([a.sub.i] * [x.sub.j] + [b.sub.j]) = [x.sub.j], j = 1, 2, ..., L. (23)

The output weight can be obtained with the following:

[mathematical expression not reproducible], (24)

where [a.sub.i] = [[[a.sub.i1], [a.sub.i2], ..., [a.sub.in]].sup.T] is the input weight generated randomly and [x.sub.i] = [[x.sub.il], [x.sub.i2], ..., [x.sub.in]] are the input and output of the OCI-ELM-AE.

5.2. Implementation of Stacked OCI-ELM Autoencoders in Deep Network. In 2006, Hinton et al. [31] presented the concept of deep learning to solve the problems of unsupervised data. Deep belief nets (DBNs) are probabilistic generative models which are first trained only with unlabeled data and then fine-tuned in a supervised mode. And then, another kind of deep network based on Restricted Boltzmann Machine (RBM) [32], deep Boltzmann machine (DBM) [33], was introduced by Salakhutdinov and Larochelle. The base building block of DBN is RBM. The ML-ELM was presented by Kasun et al. in 2013 [29]. There is no difference between the ML-ELM and the other deep learning models; ML-ELM performs layerwise unsupervised learning to train the parameters with the hidden layer weights which are initialized with ELM-AE, and the ML-ELM does not need to be fine-tuned. The AE-S-ELMs was proposed by Zhou et al. in 2014 [34]. The network consists of multiple ELMs with a small number of hidden nodes in each layer to substitute a single ELM with a large number of hidden nodes, and it implements ELM autoencoder in each iteration of S-ELMs algorithm to further improve the testing accuracy, especially for the unstructured large data without properly selected features.

Algorithm 6 (deep network based on stacked orthogonal convex incremental ELM autoencoders (DOC-IELM-AEs)). Given a training dataset N = {([x.sub.i], [t.sub.i]) | [x.sub.i] [member of] [R.sup.n], [t.sub.i] [member of] [R.sup.m], i= 1, 2, ..., N}, where [x.sub.i] = [[[x.sub.i1], [x.sub.i2], ..., [x.sub.in]].sup.T] and [t.sub.i] = [[t.sub.i1], [t.sub.in], ..., [t.sub.im]], and given activation function [sigma](x), maximum number of hidden nodes in single layer [L.sub.max], maximum number of iterations [K.sub.max], and expected learning accuracy [epsilon], one has the following.

Step 1 (initialization). Let the number of initial hidden nodes [L.sub.1[right arrow][??]] = 0, the number of iterations [K.sub.1[right arrow][??]] = 0, and residual error E = t, where t = [[[t.sub.1], t - 2, ..., [t.sub.N]].sup.T].

Step 2 (orthogonal convex I-ELM autoencoder on layer 1). This step consists of two steps as follows.

Orthogonalization Step. In this step, the following is carried out:

(a) Increase the number of hidden nodes [L.sub.1] and [K.sub.1] by one, respectively: [L.sub.1] = [L.sub.1] + 1 and [K.sub.1] = [K.sub.1] + 1.

(b) Randomly assign hidden node parameters ([a.sub.L], [b.sub.L]) for new hidden node L and calculate the output [g.sub.k],

[mathematical expression not reproducible], (25)

and the hidden layer output matrix,

[mathematical expression not reproducible]. (26)

Learning Step. While [L.sub.1] < [L.sub.max], [parallel]E[parallel] > [epsilon],

(c) calculate the output weight [[beta].sub.1(L)] for the newly added hidden node:

[beta]1(L) = E * [E - (F - [[V.sub.1(L)])].sup.T]/ E - F - [V.sub.1(L)]] * [[E - (F - V1(L))].sup.T]; (27)

(d) recalculate the output weight vectors of all existing hidden nodes if [L.sub.1] > 1:

[beta]1(i) = (1 - [beta]1(L)) [beta]11(i), I = 1, 2, ..., L - 1; (28)

(e) calculate the residual error after adding the new hidden node [L.sub.1]:

E = (1 - [beta]1(L)) E + [[beta].sub.1(L)] (F - [V.sub.1(L)]). (29)

Step 3 (orthogonal convex I-ELM autoencoder on layer 2 [right arrow] [??]). This step is carried out as follows.

Learning Step. While [L.sub.2[right arrow][??]] [L.sub.max], [parallel]E[parallel] > [epsilon],

(a) calculate the output weight [[beta].sub.1(L)] for the newly added hidden node with the hidden layer output matrix V:

[mathematical expression not reproducible]; (30)

(b) recalculate the output weight vectors of all existing hidden nodes if [L.sub.2[right arrow][??]] > 1:

[mathematical expression not reproducible]; (31)

(c) calculate the residual error after adding the new hidden node [L.sub.2[right arrow][??]]:

E = (1 - [[beta].sub.2[right arrow][??](L)])E + [[beta].sub.1] (L)(F - [[beta].sub.2[right arrow][??](L)]); (32)

Endwhile.

The DOC-IELM-AEs algorithm inherits the advantages of incremental constructive feedforward networks model and deep learning algorithms on exactly capturing higher-level abstractions and characterizing the data representations. The implementation of autoencoders for the unsupervised pretraining of data exhibits super-duper performance on regression and classification problems. The improved method utilizes the OCI-ELM-AE as abase building layer to construct the whole deep architecture. As shown in Figure 2, the data is mapped to OCI-ELM feature space; in each layer, OCI-ELM-AE output weights with respect to input data are the weight of the first layer; for the same reason, the output weights of OCI-ELM-AE, with respect to hidden layer output, are the layer weights of DOC-IELM-AEs. The detailed algorithm of DOC-IELM-AEs is shown in Algorithm 6.

5.2.1. Performance Comparison of Regression Problems Based on DOC-IELM-AEs. In this section, we mainly test the regression performance of the proposed OCI-ELM and DOC-IELM-AEs on three UCI real-world datasets, Parkinsons, California Housing, and CCS (Concrete Compressive Strength) data, and two large datasets, Blog Feedback and Online News Popularity data. The simulations are conducted in MATLAB 2013a environment running on Windows 7 machine with 128 GB of memory and Intel Xeon E5-2620V2 (2.1 GHz) processor.

The regression performance comparisons of the proposed algorithms OCI-ELM and DOC-IELM-AEs with the baseline methods including SVM [35], single ELM, ML-ELM, AE-S-ELMs, DBN, ErrCor [36], and PC-ELM are shown in Table 5. The specific analyses on the results of regression capability and effectiveness are as follows:

(1) OCI-ELM compared with SVM, ELM, ErrCor, and PC-ELM: we perform the regression testing on the datasets described in Table 4. The simulations are obtained by the average of 50 trails; we can observe from Table 4 that the testing accuracies of OCI-ELM on UCI datasets and large datasets are both better than SVM, ELM, ErrCor, and PC-ELM. For BlogFeedback dataset, the training accuracy of OCIELM is 91.76%, and those of SVM, ELM, ErrCor, and PC-ELM are 89.75%, 90.12%, 90.39%, and 90.54%, respectively. Meanwhile, OCI-ELM also obtains better testing accuracy of 91.82% than other algorithms. Although the OCI-ELM is an iterative learning algorithm, the compact neural network makes the convergence rate faster than PC-ELM and ErrCor, merely slower than SVM and ELM. Thus, the training time consumed on OCI-ELM learning is acceptable.

(2) DOC-IELM-AEs compared with DBN, ML-ELM, and AE-S-ELMs: the testing accuracy on UCI datasets can show that the performance of DOC-IELMAEs outperforms the OCI-ELM. With the aforementioned comparisons between OCI-ELM and the other algorithms (SVM, ELM, ErrCor, and PC-ELM), evidenced by the same token, the DOC-IELM-AEs can achieve better testing accuracy than SVM, ELM, ErrCor, and PC-ELM; this result can also be seen in Table 5. For the large-scale datasets (BlogFeedback and Online News Popularity), the DOC-IELM-AEs obtained accuracies of 93.16%, 93.27% and 93.69%, 93.84% for training and testing with the network structures 281-1000-1000-2000-10 and 61-700-700-10000-26, respectively. The simulations in Table 5 show that DOC-IELM-AEs can produce better results than DBN, ML-ELM, and AE-S-ELMs. Furthermore, DOC-IELM-AEs enjoys the advantage over the DBN and ML-ELM on training speed. Thus, with the better regression performance, DOC-IELM-AEs would provide the state-of-the-art method for large-scale unstructured data problems.

5.2.2. Performance Comparison of Classification Problems Based on DOC-IELM-AEs. The classification performance comparisons of the proposed algorithms OCI-ELM and DOC-IELM-AEs with the baseline methods including SVM, single ELM, ML-ELM, AE-S-ELMs, DBN, ErrCor, and PC-ELM are shown in Table 6. The specific comparisons are as follows:

(1) OCI-ELM compared with SVM, ELM, ErrCor, and PC-ELM: the simulation results are obtained by the average of 50 trails on datasets in Table 4 (from Delta Ailerons to NORB data). For the BlogFeedback, the training and testing accuracies are 91.76% and 91.82%, respectively, listed in Table 6; we can see that OCI-ELM achieves better classification accuracy than SVM, ELM, ErrCor, and PC-ELM. And the speed of learning is faster than other improved ELM algorithms, notwithstanding behind the SVM and single ELM due to the process of iteration learning.

(2) DOC-IELM-AEs compared with DBN, ML-ELM, and AE-S-ELMs: to test these anticipated effects, we used UCI datasets and large-scale datasets to acquire the results. From the experimental results, we can see that the classification accuracies of DOC-IELM-AEs are better than others obviously. Focusing on NORB, the network structure used by DOC-IELM-AEs is 2048-800-800-3000-5; the DOC-IELM-AEs obtained the best accuracies of 93.16%, 93.27% and 93.69%, 93.84% for training and testing, respectively, in all algorithms, including SVM, single ELM, ML-ELM, AE-S-ELMs, DBN, ErrCor, PC-ELM OCIELM, and DOC-IELM-AEs. Furthermore, the simulation results of other datasets also display the outstanding performance of DOC-IELM-AEs. Thus, with the better accuracy and faster speed of training, DOC-IELM-AEs can be applied in the vast majority of classification problems.

6. Case Study on Elongation Prediction of Strips

In this section, all of the experimental results for the elongation of strips prediction are presented. The annealing treatment is considered the most important process to cold rolled strips. In this process, the cold working hardening and internal stress of strips can be eliminated; the hardness of strips can be reduced; moreover, the ability of plastic deformation, stamping, and mechanical technique can be improved. Figure 3 shows the process of continuous annealing. In the furnace, the strips will pass five temperature sections, that is, preheating section (PHS), heating section (HS), slow cooling section (SS), rapid cooling section (RCS), and equalising section (ES), and three tension sections, that is, SS tension section, RCS tension section, and HS tension section. Therefore, the strips will extend or shorten with the changes of temperature and tension. Meanwhile, the surface friction coefficient and the rotational speed of the tension rolls also affect the elongation of strips, rendering the weld position unable to be tracked accurately, having a great influence on the rate of finished product and the safety of air-knife. Thus, the proposed method DOC-IELM-AEs is applied in the annealing of strips process to obtain the position information of welds. The annealing process has 12 continuous process measurements and 10 manipulated variables according to the experience and mechanism analysis.

We collect the historical records in the last 16 months which can affect the position of the welding seam, including the temperature data of 5 sections, the tension data of 3 sections, and the speed data of 11 sections. We use data of 10 months for training and the following data of 6 months for testing. The comparison results of elongation of strips prediction are shown in Figure 4. From the figures, we can see that the prediction results in the 6 months obtained based on 6 algorithms can all approximate the measured values. Although there are only little differences in the experimental results, the prediction based on DOC-IELM-AEs consistently outperforms the other methods in the comparisons.

For further investigation on the prediction capabilities of DOC-IELM-AEs, the performances of algorithms are evaluated in terms of four criteria, that is, the mean absolute percentage error (MAPE), the mean square error (MSE), the relative root-mean-square error (rRMSE), and the absolute fraction of variance ([R.sup.2]). During testing of DOC-IELM-AEs and other algorithms, they are also defined by using the following equations:

[mathematical expression not reproducible], (33)

where [[gamma].sub.i] and [[gamma]'.sub.i] are measured value and predicted value, respectively, and n is the number of testing data. The smaller MAPE, MSE, and rRMSE and the larger [R.sup.2] are indicative of better generalization performance of algorithm.

MAPE in Figure 5(a) evaluates the effect on measured values, generated by the disparity between the elongation values of steel-strips and the predicted values. Meanwhile, MSE and rRMSE in Figures 5(b) and 5(c), respectively, also reflect the dispersion of models, where MSE are more sensitive to the large errors compared with rRMSE because the squared errors amplify the large errors further. [R.sup.2] in Figure 5(d) is the distance between errors and predicted values. [R.sup.2] closer to 1 means that the algorithms have better performances. By analyzing the comparisons, it has become apparent that the results for evaluation criteria based on arbitrary 6-month testing data show better generalization performances of DOC-IELM-AEs than other algorithms in experiments of comparison. Accordingly, there is important practical significance in the prediction of elongation of steel-strips using DOC-IELM-AEs.

In order to demonstrate the effectiveness of the algorithm proposed in practical engineering, we have selected the successive data of 12 months from the whole data (16-month data) to conduct the comparisons, and we obtained the prediction accuracies of every month and one year. The comparisons of prediction accuracy shown in Figure 6 indicate that the performance of the algorithm proposed is the best overall, with mean accuracy of 96.795% in 12 months (all the year), compared with those of 92.49%, 92.71%, 94.47%, 94.62%, 94.43%, 93.18%, 93.22%, and 94.50% obtained from SVM, ELM, ML-ELM, AE-S-ELMs, DBN, ErrCor, PC-ELM, and OCI-ELM methods, respectively. DOC-IELM-AEs has the best accuracy among the nine methods, which indicates the predictive stability and performance of the method. Therefore, we can get the conclusion that DOC-IELM-AEs has the best prediction performance in the testing, and the algorithm proposed is a very effective method.

7. Conclusions

In this paper, we proposed a stacked architecture with OCI-ELM algorithm based on deep representation learning and added the OCI-ELM autoencoder into each layer of OCI-ELM, called DOC-IELM-AEs. The experiment results have demonstrated strongly that DOC-IELM-AEs can be suitable for solving regression and classification problems; simulations showed that, (1) compared with CI-ELM, EI-ELM, ECI-ELM, PC-ELM, and OCI-ELM, DOC-IELM-AEs can achieve the best testing accuracy with the same network size, even less hidden nodes; meanwhile, the speed of learning is also faster than other algorithms. Moreover, DOC-IELMAEs has better performance than OCI-ELM algorithm; (2) compared with SVM, ELM, ML-ELM, AE-S-ELMs, DBN, ErrCor, PC-ELM, and OCI-ELM, DOC-IELM-AEs can also obtain the best testing accuracy with consuming more time in a certain range for the large datasets; (3) compared with SVM, ELM, ML-ELM, AE-S-ELMs, DBN, ErrCor, PC-ELM, and OCI-ELM, the DOC-IELM-AEs applied in the case of strips-elongation prediction can enhance the performance of prediction; demonstrated with the production data, the prediction accuracy based on the algorithm we proposed outperforms other algorithms. For these reasons, the OCI-ELM and DOC-IELM-AEs can further be implemented in practical engineering and have the potential for solving more complicated big data problems with further study.

Competing Interests

The authors declare that they have no competing interests.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (61102124) and Liaoning Key Industry Programme (JH2/101).

References

[1] G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew, "Extreme learning machine: theory and applications," Neurocomputing, vol. 70, no. 1-3, pp. 489-501, 2006.

[2] G.-B. Huang, D. H. Wang, and Y. Lan, "Extreme learning machines: a survey," International Journal of Machine Learning and Cybernetics, vol. 2, no. 2, pp. 107-122, 2011.

[3] X.-Z. Wang, Q.-Y. Shao, Q. Miao, and J.-H. Zhai, "Architecture selection for networks trained with extreme learning machine using localized generalization error model," Neurocomputing, vol. 102, pp. 3-9, 2013.

[4] A. M. Fu, C. R. Dong, and L. S. Wang, "An experimental study on stability and generalization of extreme learning machines," International Journal of Machine Learning and Cybernetics, vol. 6, no. 1, pp. 129-135, 2015.

[5] X.-Z. Wang, R. A. R. Ashfaq, and A.-M. Fu, "Fuzziness based sample categorization for classifier performance improvement," Journal of Intelligent and Fuzzy Systems, vol. 29, no. 3, pp. 1185-1196, 2015.

[6] J. Wu, S. T. Wang, and F.-L. Chung, "Positive and negative fuzzy rule system, extreme learning machine and image classification," International Journal of Machine Learning and Cybernetics, vol. 2, no. 4, pp. 261-271, 2011.

[7] S. Lu, X. Wang, G. Zhang, and X. Zhou, "Effective algorithms of the Moore-Penrose inverse matrices for extreme learning machine," Intelligent Data Analysis, vol. 19, no. 4, pp. 743-760, 2015.

[8] G.-B. Huang, L. Chen, and C.-K. Siew, "Universal approximation using incremental constructive feedforward networks with random hidden nodes," IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

[9] J. Zhang, S. Ding, N. Zhang, and Z. Shi, "Incremental extreme learning machine based on deep feature embedded," International Journal of Machine Learning and Cybernetics, vol. 7, no. 1, pp. 111-120, 2016.

[10] Y. Ye and Y. Qin, "QR factorization based Incremental Extreme Learning Machine with growth of hidden nodes," Pattern Recognition Letters, vol. 65, pp. 177-183, 2015.

[11] J.-L. Ding, F. Wang, H. Sun, and L. Shang, "Improved incremental regularized extreme learning machine algorithm and its application in two-motor decoupling control," Neurocomputing, vol. 149, pp. 215-223, 2015.

[12] Z. Xu, M. Yao, Z. Wu, and W. Dai, "Incremental regularized extreme learning machine and it's enhancement," Neurocomputing, vol. 174, pp. 134-142, 2016.

[13] G.-B. Huang, M.-B. Li, L. Chen, and C.-K. Siew, "Incremental extreme learning machine with fully complex hidden nodes," Neurocomputing, vol. 71, no. 4-6, pp. 576-583, 2008.

[14] Y. Li, "Orthogonal incremental extreme learning machine for regression and multiclass classification," Neural Computing & Applications, vol. 27, no. 1, pp. 111-120, 2016.

[15] G.-B. Huang and L. Chen, "Convex incremental extreme learning machine," Neurocomputing, vol. 70, no. 16-18, pp. 3056-3062, 2007.

[16] G. E. Hinton and R. R. Salakhutdinov, "Reducing the dimensionality of data with neural networks," Science, vol. 313, no. 5786, pp. 504-507, 2006.

[17] S. Ding, N. Zhang, X. Xu, L. Guo, and J. Zhang, "Deep extreme learning machine and its application in EEG classification," Mathematical Problems in Engineering, vol. 2015, Article ID 129021,11 pages, 2015.

[18] O. Vinyals, Y. Jia, L. Deng, and T. Darrell, "Learning with recursive perceptual representations," in Advances in Neural Information Processing Systems, pp. 2825-2833, 2012.

[19] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "ImageNet classification with deep convolutional neural networks," in Advances in Neural Information Processing Systems, vol. 2, no. 25, MIT Press, 2012.

[20] R. Socher, J. Pennington, E. H. Huang, A. Y. Ng, and C. D. Manning, "Semi-supervised recursive autoencoders for predicting sentiment distributions," in Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP '11), pp. 151-161, Association for Computational Linguistics, July 2011.

[21] Y. Bengio and O. Delalleau, "On the expressive power of deep architectures," in Algorithmic Learning Theory, J. Kivinen, C. Szepesvari, E. Ukkonen, and T. Zeugmann, Eds., vol. 6925 of Lecture Notes in Computer Science, pp. 18-36, Springer, New York, NY, USA, 2011.

[22] Y. Bengio, "Learning deep architectures for AI," Foundations and Trends in Machine Learning, vol. 2, no. 1, pp. 1-27, 2009.

[23] Y. Bengio and Y. Lecun, "Scaling learning algorithms towards AI," Large-Scale Kernel Machines, vol. 2007, no. 34, pp. 1-41, 2007.

[24] T. S. Shores, Applied Linear Algebra and Matrix Analysis, Springer, Berlin, Germany, 2007.

[25] G. Taguchi and R. Jugulum, The Mahalanobis Taguchi Strategy: A Pattern Technology System, John Wiley & Sons, Hoboken, NJ, USA, 2002.

[26] Y. M. Yang, Y. N. Wang, and X. F. Yuan, "Parallel chaos search based incremental extreme learning machine," Neural Processing Letters, vol. 37, no. 3, pp. 277-301, 2013.

[27] Q. Yu, Y. Miche, E. Severin, and A. Lendasse, "Bankruptcy prediction using Extreme Learning Machine and financial expertise," Neurocomputing, vol. 128, pp. 296-302, 2014.

[28] K. I. Wong, M. V. Chi, P. K. Wong et al., "Sparse Bayesian extreme learning machine and its application to biofuel engine performance prediction," Neurocomputing, vol. 2015, no. 149, pp. 397-404, 2015.

[29] L. L. C. Kasun, H. Zhou, G. B. Huang, and C. M. Vong, "Representational learning with extreme learning machine," IEEE Intelligent Systems, vol. 6, no. 28, pp. 31-34, 2013.

[30] W. Johnson and J. Lindenstrauss, "Extensions of Lipschitz maps into a Hilbert space," Modern Analysis and Probability, vol. 189, no. 26, pp. 189-206, 1984.

[31] G. E. Hinton, S. Osindero, and Y.-W. Teh, "A fast learning algorithm for deep belief nets," Neural Computation, vol. 18, no. 7, pp. 1527-1554, 2006.

[32] G. E. Hinton, "A practical guide to training restricted Boltzmann machines," Momentum, vol. 1, no. 9, pp. 599-619, 2010.

[33] R. Salakhutdinov and H. Larochelle, "Efficient learning of deep Boltzmann machines," in Proceedings of the 13th International Conference on Artificial Intelligence and Statistics (AISTATS '10), vol. 9 of JMLR: Workshop and Conference Proceedings, pp. 693-700, 2010.

[34] H. Zhou, G. B. Huang, Z. Lin et al., "Stacked extreme learning machines," IEEE Transactions on Cybernetics, vol. 2, no. 2, pp. 1-13, 2014.

[35] M. A. Hearst, S. T. Dumais, E. Osman, J. Platt, and B. Scholkopf, "Support vector machines," IEEE Intelligent Systems, vol. 13, no. 4, pp. 18-28, 1998.

[36] H. Yu, P. D. Reiner, T. Xie, T. Bartczak, and B. M. Wilamowski, "An incremental design of radial basis function networks," IEEE Transactions on Neural Networks and Learning Systems, vol. 25, no. 10, pp. 1793-1803, 2014.

http://dx.doi.org/10.1155/2016/1649486

Chao Wang, Jianhui Wang, and Shusheng Gu

College of Information Science and Engineering, Northeastern University, Shenyang 110819, China

Correspondence should be addressed to Chao Wang; supper_king1018@163.com

Received 1 March 2016; Revised 7 May 2016; Accepted 31 May 2016

Academic Editor: Lotfi Senhadji

Caption: Figure 1: Network diagram of OCI-ELM-AE.

Caption: Figure 2: The model structure of DOC-IELM-AEs.

Caption: Figure 3: General scheme of strip steel in annealing process.

Caption: Figure 4: The comparisons of prediction on 6-month strip-elongation data.

Caption: Figure 5: Different prediction errors of 5 models on 6-month data.

Caption: Figure 6: The comparisons of prediction accuracy of annual statistical data.

Table 1: Specification of 10 benchmark problems. Training Testing Datasets sample sample Attributes Type Auto MPG 182 165 8 Regression California Housing 8,800 8,260 8 Regression Servo 92 60 4 Regression Concrete Compressive 975 830 9 Regression Strength Parkinsons 4,780 5,120 26 Regression Delta Ailerons 5,080 4,600 6 Classification Waveform II 3,300 2,800 40 Classification Abalone 3,900 3,670 8 Classification Breast Cancer 539 490 10 Classification Energy Efficiency 740 600 8 Classification Table 2: The comparisons of training and testing on the regression cases. Datasets Approaches [CI-ELM.sup.(2007)] [PC-ELM.sup.(2012)] [LOO-IELM.sup.(2014)] Auto MPG (0.08) [SB-ELM.sup.(2014)] [II-RELM.sup.(2015)] [EIR-ELM.sup.(2016)] OCI-ELM CI-ELM PC-ELM LOO-IELM California Housing (0.12) SB-ELM II-RELM EIR-ELM OCI-ELM CI-ELM PC-ELM LOO-IELM Servo (0.115) SB-ELM II-RELM EIR-ELM OCI-ELM CI-ELM PC-ELM LOO-IELM CCS (0.035) SB-ELM II-RELM EIR-ELM OCI-ELM CI-ELM PC-ELM LOO-IELM Parkinsons (0.14) SB-ELM II-RELM EIR-ELM OCI-ELM RMSE (training & testing) Nodes Datasets (fixed) Training Testing 20 0.1043 0.1035 20 0.1014 0.1012 20 0.1106 0.1104 Auto MPG (0.08) 20 0.1376 0.2307 20 0.0998 0.1005 20 0.0893 0.1005 20 0.0827 0.0823 150 0.1601 0.1583 150 0.1389 0.1377 150 0.1376 0.1374 California Housing (0.12) 150 0.1363 0.1369 150 0.1341 0.1339 150 0.1274 0.1268 150 0.1272 0.1263 100 0.1428 0.1419 100 0.1373 0.1364 100 0.1371 0.1368 Servo (0.115) 100 0.1257 0.1254 100 0.1303 0.1307 100 0.1265 0.1264 100 0.1238 0.1232 150 0.0611 0.0602 150 0.0381 0.0365 150 0.0372 0.0369 CCS (0.035) 150 0.0366 0.0368 150 0.0361 0.0363 150 0.0348 0.0351 150 0.0332 0.0346 250 0.0913 0.0906 250 0.0471 0.0463 250 0.0453 0.0462 Parkinsons (0.14) 250 0.0388 0.0391 250 0.0389 0.0383 250 0.0344 0.0347 250 0.0301 0.0283 Hidden nodes & average time Datasets # nodes Time (s) 66.29 0.1485 34.07 0.2783 49.91 0.3183 Auto MPG (0.08) =130 0.0717 44.17 0.3483 31.05 0.3283 23.62 0.2204 330.09 1.0051 199.34 0.9810 217.08 0.9766 California Housing (0.12) -- -- 192.33 0.9713 184.67 1.0017 172.15 0.9704 182.63 0.0806 160.82 0.0701 155.72 0.0765 Servo (0.115) =127 0.0355 157.12 0.0886 147.80 0.0794 143.56 0.0828 229.86 0.5893 162.79 1.1236 159.04 0.9427 CCS (0.035) =170 0.0872 163.82 0.8341 145.78 0.6305 130.04 0.5835 170.02 3.3403 63.55 4.7503 59.78 4.4452 Parkinsons (0.14) -- -- 77.19 4.4189 48.92 3.9836 39.92 3.0819 Table 3: The comparisons of training and testing on the classification cases. Datasets Approaches [CI-ELM.sup.(2007)] [PC-ELM.sup.(2012)] Delta Ailerons (0.035) [LOO-IELM.sup.(2014)] [SB-ELM.sup.(2014)] [II-RELM.sup.(2015)] [EIR-ELM.sup.(2016)] OCI-ELM CI-ELM PC-ELM Waveform II (0.04) LOO-IELM SB-ELM II-RELM EIR-ELM OCI-ELM CI-ELM PC-ELM Abalone (0.05) LOO-IELM SB-ELM II-RELM EIR-ELM OCI-ELM CI-ELM PC-ELM Breast Cancer (0.07) LOO-IELM SB-ELM II-RELM EIR-ELM OCI-ELM CI-ELM PC-ELM Energy Efficiency (0.055) LOO-IELM SB-ELM II-RELM EIR-ELM OCI-ELM Testing accuracy Datasets Nodes Mean (%) Std. (fixed) 250 83.29 0.0036 250 90.02 0.0016 Delta Ailerons (0.035) 250 91.17 0.0027 250 91.66 0.0071 250 91.18 0.0042 250 92.03 0.0019 250 92.84 0.0012 100 84.47 0.0182 100 89.81 0.0104 Waveform II (0.04) 250 88.93 0.0097 250 80.69 0.0181 250 90.64 0.0112 250 91.15 0.0096 100 93.11 0.0083 150 82.72 0.0022 150 93.57 0.0016 Abalone (0.05) 250 90.51 0.0033 250 86.03 0.0107 250 92.10 0.0034 250 93.91 0.0018 150 94.24 0.0015 200 90.06 0.0145 200 93.23 0.0082 Breast Cancer (0.07) 250 93.07 0.0104 250 94.41 0.0075 250 92.58 0.0095 250 94.76 0.0078 200 94.73 0.0067 150 91.78 0.0013 150 96.53 0.0008 Energy Efficiency (0.055) 250 95.18 0.0024 250 92.16 0.0033 250 95.29 0.0011 250 96.67 0.0012 150 97.25 0.0008 Hidden nodes & average time Datasets # nodes Time (s) 369.32 1.3505 35.19 0.6829 Delta Ailerons (0.035) 41.12 0.7761 =220 0.0556 51.16 0.7425 34.29 1.1304 31.73 0.7021 200.11 3.0977 47.63 3.0954 Waveform II (0.04) 46.44 3.3437 -- -- 44.33 3.6603 38.91 3.2267 29.54 3.0864 150.37 0.4930 24.62 0.6177 Abalone (0.05) 38.43 0.7102 =180 0.0413 35.74 0.6924 24.16 0.8533 21.41 0.6802 88.30 0.0804 34.79 0.0992 Breast Cancer (0.07) 40.82 0.1102 =150 0.0275 55.27 0.1032 31.18 0.1174 34.58 0.1061 61.09 0.2966 41.08 0.3617 Energy Efficiency (0.055) 27.94 0.3517 =150 0.0658 46.83 0.3826 22.42 0.4011 18.49 0.3979 Table 4: Specification of 10 benchmark problems. Datasets Training Testing Attributes Type sample sample Parkinsons 4,780 5,120 26 Regression California 8,800 8,260 8 Regression Housing Concrete 975 830 9 Regression Compressive Strength BlogFeedback 58,000 23,000 281 Regression Online News 36,800 10,000 61 Regression Popularity Delta Ailerons 5,080 4,600 6 Classification Waveform II 3,300 2,800 40 Classification MNIST 35,000 6,000 784 Classification OCR Letters 40,000 12,000 128 Classification NORB 20,000 5,000 2,048 Classification Table 5: The comparisons of training and testing on the regression cases. Training Training Datasets Algorithms accuracy (%) time (s) SVM 95.42 282.26 ELM 95.75 240.68 ML-ELM 97.93 482.7 AE-S-ELMs 98.07 492.21 Parkinsons DBN 97.51 5013 ErrCor 96.25 344.82 PC-ELM 97.16 362.05 OCI-ELM 97.59 327.2 DOC-IELM-AEs 98.44 440.39 SVM 96.56 327.17 ELM 96.79 289.24 ML-ELM 98.12 462.31 AE-S-ELMs 98.26 402.49 California DBN 98.03 3989 Housing ErrCor 97.01 401.66 PC-ELM 97.34 396.52 OCI-ELM 97.92 386.63 DOC-IELM-AEs 98.71 447.2 SVM 95.72 32.32 ELM 95.91 28.29 ML-ELM 97.26 53.4 AE-S-ELMs 97.49 50.66 Concrete h DBN 96.64 320.04 Compressive ErrCor 96.32 47.45 Strength PC-ELM 96.55 43.97 OCI-ELM 97.04 44.33 DOC-IELM-AEs 98.39 51.58 SVM 89.75 3906 ELM 90.12 2405 ML-ELM 91.86 5175 AE-S-ELMs 91.83 5247 BlogFeedback DBN 90.51 19766 ErrCor 90.39 4889 PC-ELM 90.54 4308 OCI-ELM 91.76 4223 DOC-IELM-AEs 93.16 5271 SVM 91.75 311.83 ELM 91.68 131.34 ML-ELM 92.62 684.98 AE-S-ELMs 92.59 685.62 Online News DBN 92.23 7062 Popularity ErrCor 91.61 541.14 PC-ELM 92.29 576.01 OCI-ELM 92.52 521.24 DOC-IELM-AEs 93.69 634.09 Testing Testing Datasets accuracy (%) deviation (%) 95.43 0.01 95.73 0.29 97.96 0.08 98.05 0.11 Parkinsons 97.53 0.04 96.19 0.32 97.18 0.09 97.62 0.05 98.41 0.04 96.56 0 96.72 0.16 98.14 0.12 98.23 0.09 California 98.04 0.08 Housing 97.11 0.25 97.31 0.28 97.90 0.11 98.72 0.07 95.78 0.04 96.02 0.06 97.33 0.18 97.41 0.12 Concrete 96.58 0.05 Compressive 96.35 0.25 Strength 96.67 0.09 97.15 0.06 98.37 0.03 89.82 0.03 90.14 0.22 91.82 0.12 91.79 0.13 BlogFeedback 90.60 0.09 90.44 0.09 90.58 0.12 91.82 0.09 93.27 0.07 91.72 0.04 91.69 0.34 92.71 0.17 95.72 0.12 Online News 92.26 0.05 Popularity 91.77 0.29 92.35 0.15 92.54 0.12 93.84 0.11 Table 6: The comparisons of training and testing on the classification cases. Training Training Datasets Algorithms accuracy (%) time (s) SVM 95.41 219.54 ELM 95.78 43.88 ML-ELM 97.32 317.83 AE-S-ELMs 97.75 279.32 Delta Ailerons DBN 98.34 4432 ErrCor 96.42 234.27 PC-ELM 96.56 262.69 OCI-ELM 96.73 249.89 DOC-IELM-AEs 98.89 350.17 SVM 95.45 247.11 ELM 95.71 107.14 ML-ELM 97.83 304.28 AE-S-ELMs 98.09 365.83 Waveform II DBN 97.59 4101 ErrCor 96.54 261.44 PC-ELM 96.46 293.85 OCI-ELM 97.33 202.08 DOC-IELM-AEs 98.86 332.58 SVM 95.01 2108 ELM 95.33 1114 ML-ELM 96.78 3793 AE-S-ELMs 96.93 3772 MNIST DBN 96.67 14117 ErrCor 96.17 4065 PC-ELM 96.21 3049 OCI-ELM 96.38 3092 DOC-IELM-AEs 97.89 3985 SVM 88.27 770.24 ELM 89.02 265.01 ML-ELM 89.19 923.23 AE-S-ELMs 89.44 960.06 OCR Letters DBN 88.98 15212 ErrCor 88.81 1014 PC-ELM 89.17 979.62 OCI-ELM 89.36 992.08 DOC-IELM-AEs 90.85 1175 SVM 91.13 723.81 ELM 91.42 66.07 ML-ELM 92.67 1107 AE-S-ELMs 92.81 698.21 NORB DBN 92.58 44502 ErrCor 91.77 1287 PC-ELM 91.80 896.63 OCI-ELM 92.61 839.48 DOC-IELM-AEs 94.53 1480 Testing Testing Datasets accuracy (%) deviation (%) 95.42 0 95.81 0.22 97.34 0.19 97.71 0.27 Delta Ailerons 97.35 0.05 96.48 0.12 96.61 0.08 96.69 0.07 98.90 0.05 95.44 0.01 95.73 0.16 97.86 0.11 98.11 0.15 Waveform II 97.52 0.04 96.58 0.06 96.47 0.05 97.36 0.08 98.85 0.05 95.03 0.03 95.32 0.31 96.82 0.17 96.94 0.21 MNIST 96.72 0.03 96.26 0.13 96.24 0.09 96.39 0.09 97.94 0.06 88.41 0.03 89.03 0.22 89.17 0.14 89.46 0.19 OCR Letters 89.02 0.04 88.82 0.10 89.16 0.05 89.39 0.08 90.86 0.06 91.14 0.04 91.45 0.14 92.70 0.17 92.82 0.12 NORB 92.59 0.03 91.81 0.08 91.83 0.06 92.65 0.05 94.56 0.03

Printer friendly Cite/link Email Feedback | |

Title Annotation: | Research Article; Extreme learning machine |
---|---|

Author: | Wang, Chao; Wang, Jianhui; Gu, Shusheng |

Publication: | Mathematical Problems in Engineering |

Article Type: | Report |

Geographic Code: | 1USA |

Date: | Jan 1, 2016 |

Words: | 9571 |

Previous Article: | Shafting Alignment Computing Method of New Multibearing Rotor System under Specific Installation Requirement. |

Next Article: | Schedule-Based Passenger Assignment for High-Speed Rail Networks considering the Ticket-Booking Process. |

Topics: |