Cognitive Computational Model Using Machine Learning Algorithm in Artificial Intelligence Environment

Machine learning includes three forms: supervised learning, unsupervised learning and reinforcement learning [13]. Algorithms related to machine learning include inverse linear regression, logical regression, naïve Bayes, K-nearest neighbour algorithm, support vector machine (SVM), random forest, boosting and AdaBoost [14]. Its advantage is that the same algorithm can be used to analyse and solve many different things, and the process and operation are relatively simple [15].

2.1.1

DBN algorithm

First, we consider the algorithm function. The DBN algorithm belongs to a kind of neural network in machine learning and can have both unsupervised learning and supervised learning functions [16]. DBN is a probability generation model, which establishes a combination distribution between observation data and tags. By training the weights between units, the neural network can train the data information with the maximum probability as the goal. DBN is used for feature recognition, data classification and probability generation. DBN algorithm has strong practicability, wide application fields and high expansion performance. It can be used in handwriting recognition, speech recognition and image processing in machine learning [17].

Second, we look at the structure and principle of the algorithm. The components of DBN are multilayer units (visible units and hidden units). Visible units can be used to receive input, and hidden units can be used for feature extraction [18]. The connection between the top two layers of units has no direction, forming the associative memory. The other neural layers belong to directed connection, and the bottom layer belongs to data vectors. A unit represents a dimension of the data vector. Further, the restricted Boltzmann machine (RBM) is a component of DBN, which can train the DBN layer by layer. The data vector can be used to judge the hidden layer in each layer. Then, this layer is used as the data vector of the unit in the next higher layer. The RBM can function independently as a clusterer. There are only two layers of units in RBM, one of which is the visible layer for training input data. The other layer is the hidden layer, which plays the function of feature detector. Figure 1 shows its structure.

Fig. 1

In Figure 1, the units (grey boxes) on the left form the hidden layer, and the units on the right form the visible layer. Each layer can be represented by a vector; each unit represents a dimension, and bidirectional connection is carried out between two layers.

In order to make the research content more general, each layer of the DBN algorithm training is a Markov random field model, and each model contains a visible layer and a hidden layer. U = (U₁,...,U_n) represents the variable matrix in the visible layer, P = (P₁,P_n) represents the variable matrix of the hidden layer. Based on (U, P) Gibbs distribution, the variable matrix P of the hidden layer is used to describe the marginal distribution of the variable matrix U of the visible layer. (1) S(U)=∑mS(U,P)=1K∑mexp(−∈(U,P)). S\left(U \right) = \sum\limits_m {S\left({U,P} \right) = \frac{1}{K}}\sum\limits_{\rm{m}} {\exp \left({- \in \left({U,P} \right)} \right)}.

In Eq. (1), K=∑U,Pexp(−∈(U,P)) K = \sum\limits_{U,P} {\exp \left({- \in \left({U,P} \right)}\right)} . Therefore, the likelihood function of parameter λ can be defined as follows: (2) lnL(λ|U)=lnS(λ|U)=ln∑Pexp(−∈(U,P))−ln∑U,Pexp(−∈(U,P)). \ln L\left(\lambda \left| U \right. \right) = \ln S\left(\lambda \left| U\right. \right) = \ln \sum\limits_{P} \exp \left({- \in \left({U,P} \right)}\right) - \ln \sum\limits_{U,P} \exp \left({- \in \left({U,P} \right)}\right).

According to Eq. (2), the gradient of parameter attribute adjustment is calculated. (3) αlnL(λ|U)αλ=−ln∑PP(P|U)∂∈(U,P)αλ+ln∑PP(P,U)∂∈(U,P)αλ. \frac{{\alpha \ln L\left(\lambda \left| U \right. \right)}}{{\alpha \lambda}} = - \ln \sum\limits_P {P\left(P\left| U \right. \right)} \frac{{\partial \in \left({U,P} \right)}}{{\alpha \lambda}} + \ln \sum\limits_P {P\left({P,U}\right)} \frac{{\partial \in \left({U,P} \right)}}{{\alpha \lambda}}.

Here, α represents the parameter adjustment value.

Each RBM contains a hidden layer and a visible layer, which represent different variables. The visible and hidden layers are not connected with itself. If the values are 0 and 1, the energy equation of the RBM can be obtained. (4) E(U,P)=−∑x=1a∑y=1bUxOxyPy−∑x=1bIxUx−∑y=1aRyPy. E\left({U,P} \right) = - \sum\limits_{x = 1}^a {\sum\limits_{y = 1}^b {{U_x}}} {O_{xy}}P{}_y - \sum\limits_{x = 1}^b {{I_x}} {U_x} - \sum\limits_{y = 1}^a{{R_y}{P_y}}.

Here, O, I and R represent the paranoid coefficients, and a and b represent the maximum ranges of the paranoid coefficient.

According to Eq. (4), it can be identified that in the energy equation, the weight should be adjusted globally and the bias coefficient should be relatively independent. If RBM is regarded as an image, the image only contains the connection between the hidden layer and the visible layer, and there is no connection inside the layer. According to Gibbs distribution and Eq. (4), the conditional probability equation can be obtained. (5) p(Uy=1|P)=σ(∑y=1aOxyPy+Iy)p(Py=1|U)=σ(∑x=1aOxyUy+Rx) p\left({U_y} = 1\left| P \right. \right) = \sigma \left({\sum\limits_{y =1}^a {{O_{xy}}P{}_y + {I_y}}} \right)\\ \nonumber p\left({P_y} = 1\left| U \right. \right) = \sigma \left({\sum\limits_{x =1}^a {{O_{xy}}U{}_y + {R_x}}} \right) where p is the probability and σ is the coefficient.

Then, the RBM network is trained using the contrastive divergence algorithm. After the parameter of paranoid ideation is adjusted, the learning function of parameter λ can be obtained. (6) λT+1=φ⋅λT+μ⋅{ααλ[1V∑Z=1VlnL(λ|Uy)]−β⋅λT}, {\lambda^{T + 1}} = \varphi \cdot {\lambda^{T}} + \mu \cdot \left\{{\frac{\alpha}{{\alpha \lambda}}\left[ \frac{1}{V}\sum\limits_{Z = 1}^{V}{\text{lnL}} \left(\lambda \left| U_{y} \right. \right) \right] - \beta \cdot {\lambda ^T}} \right\}, where φ represents inertia factor, μ represents the learning rate, and β represents the penalty factor.

The RBM training process aims to calculate the probability set that can generate many training samples. In this probability set, the training samples have the largest probability. The decisive factor of the probability combination is the weight size, so the purpose of RBM training is to find the optimal weight.

Third, we move on DBN training. DBN is a neural network based on RBM, which can be either a generation model or a discriminant model. The process of training the DBN involves the use of unsupervised layer-by-layer analysis to get the weights. Thus, the training process of DBN should be carried out layer by layer. In each layer, the hidden layer prediction should be carried out according to the data vector. Then, the hidden layer is used as the information vector for adjacent higher layers. In other words, many RBMs can be combined to form a DBN. The visible layer (input value) of the next RBM is the hidden layer (output value) of the previous RBM. During the training process, the RBM of the previous layer must be fully trained before the RBM of the next layer can be trained, which goes all the way to the end layer. Finally, the units are the essential components of the neural network, and the DBN is composed of many units and RBMs. The restricted layer of the DBN network structure comprises the visible layer and the hidden layer, and there are corresponding connections between layers. However, there is no connection between the neurons in the interior of the layer (grey boxes). By training the units of the hidden layer, the association between the higher-order data displayed in the visible layer can be mined. Figure 2 shows the composition of a DBN network.

Fig. 2

Fourth, we come to the essence of the DBN algorithm. From the perspective of unsupervised learning, the goal is to preserve the nature of the original features to the greatest extent and to reduce the dimensions of features. From the perspective of supervised learning, the goal is to keep the error rate of classification as low as possible. Irrespective of whether supervised or unsupervised learning is adopted, the DBN algorithm is essentially a process of feature learning, in other words, a process to get the best feature expression.

Fifth, we detail the DBN training process. The first RBM needs to be fully trained; its weight and offset are fixed, and its hidden units are used as the import vector of the next RBM; then, after the second RBM is fully trained, the second RBM is overlapped with the first RBM. It is necessary to repeat the above steps several times. If the data in the training set contain labels, when the top RBM is trained, the visible layers in the RBM must contain both visible units and the units representing classification labels. Both of them are trained at the same time, as shown in Figure 3 below.

Fig. 3

For each training data, the corresponding labelled units can be set to 1 after being opened, and the remaining ones will be closed and set to 0. The parts labelled M₁ and M₂ in Figure 3 are the top labels for training in RBM.

The neural unit at the top of the network evaluates and classifies tasks based on the data generated by the lower network. When the DBN is trained at the end of the network layer, it can use the feedforward neural network to slightly adjust the previous evaluation and classification according to the data information with labels. However, the processing method of this study is better than the direct use of a feedforward neural network, because the DBN has higher efficiency, and it only changes the weight parameters and then trains in a small area. Moreover, the training speed is very fast and the convergence time is relatively short.

The DBN algorithm and multilayer perceptron are combined, and the cognitive computational model is designed. The last layer of the DBN is the output value of the RBM, which is used as the input value of the multilayer perceptron. Based on the input values and training samples, the linear perceptron training is carried out to obtain the decision model. In the decision model, the conclusion whether the data information is valid or not is obtained. After the DBN training results are obtained, the multilayer perceptron training is carried out. For the training of the multilayer perceptron, an error control strategy is added. This strategy is also an important part of the joint algorithm. The input value of the multilayer perceptron is set as C = (C₁,C₂,...,C_n), the relative weight vector is set as γ = (γ₁, γ₂,...,γ_n) and the threshold is represented by η; the cognitive strategy function can be expressed as follows: (7) Y=γdX+η. Y = {\gamma^{d}}X + \eta.

If y ≥ 0, the result of the strategy is valid. If the result is <0, the policy result is invalid.

Then, the training rules of the model weight are established. (8) Q(l+1)=Ql+κ(R−Y)⋅XQ, {Q_{\left({l + 1} \right)}} = Q_{l} + \kappa \left({R - Y} \right) \cdot {X_Q}, where κ represents the learning rate parameter.

Thus, Figure 4 is a cognitive computational model based on machine learning algorithm.

Fig. 4

First, let us see the data acquisition layer. It senses the corresponding Internet data flow generated by users using computers and enjoying related services. The relative expansion of hardware functions, such as wearable devices or smart phones and other related intelligent terminals, enables the system to obtain a large volume of information. The data acquisition layer can transfer all the collected data information to the perceptual data storage layer. At the same time, in order to make the acquired perceptual data effective and reduce the excessive extra cost caused by storing invalid data information, a simple filtering strategy is set up in the data acquisition layer. Based on the properties of the data, the information is filtered to find out whether the information is the perceptual data related to the follow-up experiment. If it is, the data are passed to the data-computing layer. If it is not, it is necessary to filter out the data. Under this operation, many useless and damaged data information will not continue to occupy the system memory. Then, the filtering operation is implemented from the specified range of data information. For example, if the ambient temperature is −100 °C, it can be judged that there is an error in the measurement of this data information, and the data should be filtered. This kind of data-filtering strategy can greatly reduce the extra cost of the system.

Second is the data storage layer. The data brought in by a lot of hardware and associated services are generally massive. If these data are stored, most of the storage space of the system will be occupied. Therefore, how to find a more efficient and secure method to store and process these data is an urgent problem to be solved. Therefore, the storage requirements of the storage layer are deeply optimised, i.e., we propose a data storage layer that can carry out data perception before and after. In this layer, a distributed storage system based on different attributes of the data is proposed. In the system, based on the data classifier, the obtained data are divided into static and dynamic data, and different processing methods are formulated according to the different data types. For the static dataset (time or work attributes), the distributed database is first imported and, at the same time, a static data export port is developed for the execution of the function of the data-computing layer. For the data stream with a form of perception, such as the ambient temperature in different time periods, the performance of this kind of data has strong timeliness. If the data can be analysed in the shortest time, valuable analysis of the data can be carried out in the shortest time.

Finally, for the above types of data, the processing method of this study is to transfer from the cache space directly to the perceptual computing layer, which is provided to the upper-level computing for use. However, this will cause a different problem, and the data information that is being cached will be lost quickly. For this problem, a kind of fault-tolerant recovery strategy is formulated to store a backup while the data are used, so as to ensure the fault-tolerant performance of the data. Based on this, even if the system runs abnormally, it can still export the data that has not been analysed from the backup for research and calculation after the system is repaired. The research measures of static data are very mature at present. Therefore, in the data-computing layer, the core of the research is to process the perceptual dataset efficiently.

MATLAB software was used in the algorithm simulation of this research. The simulation part consists of 16 imported values, upper and lower environment sets and two types of export results. In the simulation, four kinds of input and four kinds of information of upper and lower layers are used, and the machine learning algorithm proposed in the research is used to process the same. The result is whether the imported data information is valid.

The import values are set to a₁, a₂, a₃ and a₄, and the upper- and lower-level data are represented by g₁, g₂, g₃ and g₄. When the corresponding value is equal to one, it represents the existence of this input value, or it can be the upper- and lower-level properties in the current environment. The context of the imported value is shown in Table 1.

In the simulation process, the machine learning algorithm is trained by using three of the 16 derived results. However, all 16 possible data points should be used for testing. The training dataset is composed of eight derived results and 200 copies. The order of these data is disordered and used in 45 iterations. The 1,600 (8 × 200) units of data are divided into 16 datasets, and each set contains 100 data points. In each iteration, all the datasets are used. At the same time, other parameters, such as the DBN, the DBN algorithm and the number of units contained in the multilayer perceptron, should be considered.

It is considered that the number of repeated runs of the training set will have a certain impact on the accuracy of the model results. The number of units in the hidden layer is set to 50, and the number of network layers is three. From 100 to 450, the training set is optimised repeatedly every 40 increments. Figure 5 shows the results.

Fig. 5

Figure 5 shows that the accuracy rate of the results obtained by the DBN algorithm combined with multilayer perceptron training according to the specified parameters is better than that of the DBN algorithm under the same parameters. Especially when the repetition number is 220, the error rate of the model is the lowest and <0.05.

When the influence of the number of hidden layer units on the accuracy of the model results is analysed, the number of iterations of the training set is 250 and the number of network layers is three. From 20 to 50, the number of units in the hidden layer is optimised every 10 increasing units. Figure 6 shows the corresponding results obtained.

Fig. 6

Figure 6 shows that when the number of units is >40, the accuracy rate of the DBN algorithm combined with multilayer perceptron is significantly higher than that of the DBN algorithm alone. Especially if the number of units is 30, the DBN algorithm combined with multilayer perceptron can give the best effect and can control the error rate to <0.05, which the DBN algorithm alone cannot obtain.

Therefore, the number of neurons in the hidden layer of the DBN network plays an important role in the accuracy of the model. When the number of neurons exceeds a certain range, the accuracy rates of the DBN network and the multilayer perceptron are basically the same. However, the accuracy rate of the combined algorithm can reach the best state, which is the effect that the DBN network and multilayer perceptron cannot achieve alone.

The number of repetitions of the training set is 300. The number of units in the hidden layer is 50. The number of network layers is changed every one increment from one to six. Each experiment is repeated 10 times, and the average value is obtained. Figure 7 shows the result.

Fig. 7

Figure 7 shows that the decision accuracy of the proposed DBN algorithm combined with multilayer perceptron is higher than that of the traditional DBN algorithm on the condition that the number of network layers in the model is equal. Especially when the number of network layers is four, the accuracy of the DBN algorithm combined with multilayer perceptron is the highest. It shows that the theoretical decision accuracy of the DBN algorithm combined with multilayer perceptron can exceed 99% under the parameters specified in the study.

There are two problems when a multilayer perceptron is used. One is that the efficiency of getting the optimal solution in the training process is not high, and it is easy to fall into the local optimal solution; the other is that the accuracy of data processing is not high. The DBN algorithm combined with the principle of multilayer perceptron can solve the above problems, and the accuracy of the results can be the highest.