CN111079843A - Training method based on RBF neural network - Google Patents
Training method based on RBF neural network Download PDFInfo
- Publication number
- CN111079843A CN111079843A CN201911309763.3A CN201911309763A CN111079843A CN 111079843 A CN111079843 A CN 111079843A CN 201911309763 A CN201911309763 A CN 201911309763A CN 111079843 A CN111079843 A CN 111079843A
- Authority
- CN
- China
- Prior art keywords
- training
- rbf neural
- neural network
- subset
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides a training method based on a RBF neural network, which belongs to the field of data classification and comprises the following steps: extracting the characteristics of each training data in the training set; dividing the characteristics of each training data in the training set into Q characteristic subsets, wherein Q is greater than 1, and the number of the characteristics contained in each characteristic subset is not less than 1; training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks; splicing the Q RBF neural sub-networks in parallel to form an RBF neural network; and training the RBF neural network based on each training data and the preset classification of each training data to obtain the trained RBF neural network. The method and the device break through the upper limit of the number of the nodes of the hidden layer of the RBF neural network, and achieve higher data classification capability.
Description
Technical Field
The application belongs to the field of data classification, and particularly relates to a training method based on a RBF neural network.
Background
With the development of modern information technology and the rapid increase of data scale, the identification capability of the traditional machine learning data classification algorithm is difficult to deal with; although the emerging deep learning algorithm has high recognition accuracy, the method also has the defects of difficult determination of a model structure, more training resource consumption, poor real-time performance and the like. Therefore, in many data classification scenarios, a classification algorithm with the advantages of simple model structure, fast real-time response, high precision and the like is needed.
The RBF neural network is derived from a radial basis function interpolation technology, only comprises an implicit layer, and adopts a nonlinear feedforward neural network with a radial basis function as an activation function of neurons of the implicit layer and a linear function as an output layer. Under certain conditions, the RBF neural network not only has the characteristics of approaching to any smooth input and output mapping and converging to global optimum, but also has the advantages of simple structure and high training speed. Within a certain range, the larger the number of nodes of the hidden layer is, the stronger the learning fitting capacity and the prediction generalization capacity of the RBF neural network are. However, the existing training algorithm of the RBF neural network processes parameters such as input features, center and width of hidden nodes, and weights of the hidden nodes and output nodes as a whole, so that the number of nodes in a hidden layer is limited, and the recognition capability of large-scale data is limited.
Disclosure of Invention
The application aims to provide a training method based on an RBF neural network, which can break through the upper limit of the number of hidden layer nodes of the RBF neural network and enable the RBF neural network to realize higher data classification capability by using more hidden layer nodes.
In order to achieve the above object, a first aspect of the present application provides a training method based on an RBF neural network, including:
extracting the characteristics of each training data in the training set;
dividing the features of each training data in the training set into Q feature subsets, wherein Q is greater than 1, and the number of features contained in each feature subset is not less than 1;
training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subset comprises and only comprises all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same; (ii) a
Splicing the Q RBF neural sub-networks in parallel to form an RBF neural network; and training the RBF neural network based on the training data and the preset classification of the training data to obtain the trained RBF neural network.
Based on the first aspect of the present application, in a first possible implementation manner, the dividing the features of each piece of training data in the training set into Q feature subsets includes:
and dividing the characteristics of each training data in the training set into Q characteristic subsets according to a rule, wherein the rule is any one of a clustering division rule, a random division rule and a sequential division rule.
Based on the first aspect of the present application or the first possible implementation manner of the first aspect of the present application, in a second possible implementation manner, the training the corresponding RBF neural sub-network based on each feature subset and the target training subset corresponding to each feature subset respectively includes:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset;
determining a target training subset corresponding to each feature subset by each feature subset;
and respectively training the RBF neural sub-networks corresponding to the feature subsets based on the number of hidden layer nodes of the feature subsets and the target training subsets.
Based on the second possible implementation manner of the first aspect of the present application, in a third possible implementation manner, the calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset respectively specifically includes:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset based on a hidden layer node calculation formula, wherein the hidden layer node calculation formula specifically comprises the following steps:
wherein the content of the first and second substances,
is to take the integer operator, N'iIs the number of features contained in the ith feature subset, N is the number of features contained in the training set, and K is the number of target hidden layer nodes, K ', of the RBF neural network'iThe number of hidden layer nodes of the RBF neural sub-network.
Based on the first aspect of the present application or the first possible implementation manner of the first aspect of the present application, in a fourth possible implementation manner, the training the RBF neural network based on the training data and the preset classification of the training data includes:
inputting each training data in the training set into an input layer of the RBF neural network, and calculating an output matrix of hidden layer nodes of the RBF neural network;
and calculating the weight from the hidden layer to the output layer in the RBF neural network based on the output matrix and the preset classification of each training data in the training set.
Based on the first aspect of the present application or the first possible implementation manner of the first aspect of the present application, in a fifth possible implementation manner, before the training the RBF neural network based on the training data and the preset classification of the training data, the method further includes:
updating the current parameters of the RBF neural network based on a gradient descent method, wherein the current parameters of the RBF neural network are formed by the parameters of each RBF neural sub-network;
the training of the RBF neural network based on the training data and the predetermined classification of the training data specifically includes:
and training the updated RBF neural network based on the training data and the preset classification of the training data.
The second aspect of the present application provides a training apparatus based on RBF neural network, including:
the extraction module is used for extracting the characteristics of each training data in the training set;
a grouping module, configured to divide the features of each training data in the training set into Q feature subsets, where Q is greater than 1, and the number of features included in each feature subset is not less than 1;
the first training module is used for training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subsets comprise and only comprise all features in the corresponding feature subsets, and input data and output data of the RBF neural sub-networks are the same;
the integrated module is used for splicing the Q RBF neural sub-networks into an RBF neural network in parallel;
and the second training module is used for training the RBF neural network based on the training data and the preset classification of the training data to obtain the trained RBF neural network.
Based on the second aspect of the present application, in a first possible implementation manner, the first training module is specifically configured to:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset;
determining a target training subset corresponding to each feature subset by each feature subset;
and respectively training the RBF neural sub-networks corresponding to the feature subsets based on the number of hidden layer nodes of the feature subsets and the target training subsets.
The third aspect of the present application provides a training apparatus based on an RBF neural network, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the first aspect or any possible implementation manner of the first aspect when executing the computer program.
A fourth aspect of the present application provides a computer-readable storage medium, in which a computer program is stored, which, when executed by a processor, performs the steps of the first aspect or any of the possible implementations of the first aspect.
As can be seen from the above, the present application first extracts the features of each training data in the training set; dividing the characteristics of each training data in the training set into Q characteristic subsets, wherein Q is greater than 1, and the number of the characteristics contained in each characteristic subset is not less than 1; training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subset comprises and only comprises all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same; splicing the Q RBF neural sub-networks in parallel to form an RBF neural network; and training the RBF neural network based on each training data and the preset classification of each training data to obtain the trained RBF neural network. According to the method, the characteristics of training data are grouped, then corresponding RBF neural sub-networks are respectively trained, and finally, all RBF neural sub-networks are spliced in parallel, so that a network structure wider than a hidden layer of a traditional RBF neural network is obtained, the upper limit of the number of nodes of the hidden layer of the traditional RBF neural network is broken through, the RBF neural network can realize higher data classification capacity by means of more numbers of the nodes of the hidden layer and stronger data fitting and generalization capacity.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed for the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without inventive work.
Fig. 1 is a schematic flowchart of a training method based on an RBF neural network according to an embodiment of the present disclosure;
fig. 2 is a schematic structural diagram of an RBF neural network-based training device according to another embodiment of the present application;
fig. 3 is a schematic structural diagram of a training apparatus based on an RBF neural network according to another embodiment of the present application.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.
It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in the specification of the present application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application, but the present application may be practiced in other ways than those described herein, and it will be apparent to those of ordinary skill in the art that the present application is not limited by the specific embodiments disclosed below.
Example one
The embodiment of the application provides a training method of an RBF neural network, which comprises the following steps:
step 11: extracting the characteristics of each training data in the training set;
optionally, the training set is a data set with M rows and N columns, where N is a number of features, M is a number of data corresponding to a certain feature, and both M and N are greater than 1; the training set may be a data set in other forms as long as the training set includes training data and corresponding features, and is not limited herein.
Specifically, all the features included in the training set (i.e., the features of each training data) are extracted, so that the extracted features are divided according to rules in step 12.
Step 12: dividing the characteristics of each training data in the training set into Q characteristic subsets;
wherein, Q is larger than 1, and the number of the characteristics contained in each characteristic subset is not less than 1;
specifically, Q is a specific natural number, and all features in the training set are divided into Q groups to obtain Q feature subsets.
Optionally, all the features in the training set are divided into Q feature subsets according to a rule, where the rule is any one of a clustering division rule, a random division rule, and a sequential division rule.
Specifically, the cluster partition rule specifically includes: setting the number of clustering centers as Q, clustering all the features in the training set into Q groups by using a clustering algorithm, and forming each feature subset by all the features belonging to each clustering center, wherein the clustering algorithm can be a k-means algorithm or other clustering algorithms, and is not limited herein.
Specifically, the random partition rule creates Q feature subsets, and sequentially selects features for each feature subset in order, so that the scale of each feature subset is the same, and the specific operation steps are as follows: according to the formula
And i-1, …, and Q, calculating the number of the features contained in the ith feature subset, wherein N is the number of the features contained in the training set, N'iThe number of features included in the ith subset of features,
for the integer operator, the features included in the ith feature subset are respectively: t is(i-1)×Q+1，T(i-1)×Q+2，T(i-1)×Q+3，…,T(i-1)×Q+QWherein T is(i-1)×Q+QRepresents the (i-1) × Q + Q features in the set of all features included in the training set.
Specifically, the sequential partitioning rule creates Q feature subsets, and randomly selects features for each feature subset in sequence, so that the scale of each feature subset is the same, and the specific operation steps are as follows: according to the formula
And i-1, …, and Q, calculating the number of the features contained in the ith feature subset, wherein N is the number of the features contained in the training set, N'iThe number of features included in the ith subset of features,
for the integer operator, the generation rule of the jth feature included in the ith feature subset is: in the interval [1, N]Randomly generating a positive integer P, checking whether the positive integer P exists in other feature subsets, and if so, re-locating in the interval [1, N ]]Randomly generating a positive integer P' and checking the positive integer P again' Presence or non-Presence in other feature subsets, repeat the above process until in the interval [1, N]If the randomly generated positive integer P "does not exist in the other feature subsets, the P" th feature in the training set is added to the ith feature subset as the jth feature.
Step 13: training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training sub-network comprise and only comprise all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same; (ii) a
Optionally, the training the corresponding RBF neural sub-network based on each feature subset and the target training subset corresponding to each feature subset respectively includes: respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset; determining a target training subset corresponding to each feature subset by each feature subset; and respectively training the RBF neural sub-networks corresponding to the feature subsets based on the number of hidden layer nodes of the feature subsets and the target training subsets.
Specifically, the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset is calculated based on a hidden layer node calculation formula, where the hidden layer node calculation formula specifically is:
wherein the content of the first and second substances,
is to take the integer operator, N'iIs the number of features included in the ith feature subset, N is the number of features included in the training set, K is the number of target hidden layer nodes, K 'of the RBF neural network'iThe number of hidden layer nodes of the ith RBF neural sub-network.
Specifically, the target training corresponding to each feature subset is determined by each feature subsetThe training set comprises the following concrete steps: let T bejExtracting the training set and the feature T for the jth feature in the ith feature subsetjCorresponding data columns form the jth column of the ith target training subset corresponding to the ith feature subset, and the steps are repeated, namely the data columns corresponding to the features in the ith feature subset in the training set are extracted to form the ith target training subset corresponding to the ith feature subset; the final target training subset obtained was M × N'iWherein M is the number of data, N ', corresponding to a certain feature in the training set'iThe number of features included in the ith feature subset.
Specifically, the training is performed by using the ith target training subset as training data of the ith RBF neural sub-network (i.e., the RBF neural sub-network corresponding to the ith feature subset), where the training process specifically includes: k'iFor the number of hidden layer nodes of the ith RBF neural sub-network, taking each piece of data in the ith target training subset as input data and output data of the ith RFB neural sub-network (namely, making the input data and the output data of the RBF neural sub-network the same), and calculating parameters of the ith RBF neural sub-network by adopting a two-stage method, wherein the parameters of the ith RBF neural sub-network comprise the center and the width of the hidden layer nodes of the RBF neural sub-network, the weights of the hidden layer nodes and the output layer nodes, and the like. Optionally, the training process for training the corresponding RBF neural sub-networks may be performed in parallel, that is, the training on the Q RBF neural sub-networks may be performed simultaneously, so as to achieve the purposes of shortening the time and increasing the training speed.
Step 14: splicing the Q RBF neural sub-networks in parallel to form an RBF neural network;
arranging the Q trained RBF neural sub-networks obtained in step 13, and splicing the Q trained RBF neural sub-networks into a wider RBF neural network in parallel, wherein the Q trained RBF neural sub-networks may be arranged sequentially or randomly, which is not limited herein.
Step 15: and training the RBF neural network based on the training data and the preset classification of the training data to obtain the trained RBF neural network.
Optionally, inputting each training data in the training set into an input layer of the RBF neural network, propagating the data forward, and calculating to obtain an output matrix of a hidden layer node of the RBF neural network; and calculating the weight from the hidden layer to the output layer in the RBF neural network based on the output matrix and the preset classification of each training data in the training set, specifically, when calculating the weight from the hidden layer to the output layer in the RBF neural network, taking the preset classification of each training data in the training set as the output value of the RBF neural network, and meanwhile, calculating the weight in a ridge regression mode according to the output matrix of the hidden layer node of the RBF neural network.
Optionally, before the RBF neural network is trained based on the training data and the preset classification of the training data, current parameters of the RBF neural network may be updated based on a gradient descent method, where the current parameters of the RBF neural network are formed by parameters of the RBF neural network, specifically, the current parameters of the RBF neural network are used as initial values, each piece of data of the training set is used as input data and output data of the RBF neural network (that is, the input data and the output data of the RBF neural network are the same), and the current parameters of the RBF neural network are updated in a fine tuning manner by using a gradient descent method.
Further, after updating the current parameters of the RBF neural network, training the RBF neural network based on the training data and the preset classification of the training data specifically includes: and training the updated RBF neural network based on the training data and the preset classification of the training data.
Based on the steps 11 to 15, a trained RBF neural network, namely a target RBF neural network with a wider hidden layer, can be obtained, the data input mode of the target RBF neural network is determined by each feature subset, the center and the width of the hidden layer node are provided by the RBF neural network formed by splicing the RBF neural sub-networks in parallel, and the weight from the hidden layer to the output layer is provided by the weight of the trained RBF neural network.
As can be seen from the above, the present application first extracts the features of each training data in the training set; dividing the characteristics of each training data in the training set into Q characteristic subsets, wherein Q is greater than 1, and the number of the characteristics contained in each characteristic subset is not less than 1; training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subset comprises and only comprises all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same; splicing the Q RBF neural sub-networks in parallel to form an RBF neural network; and training the RBF neural network based on each training data and the preset classification of each training data to obtain the trained RBF neural network. According to the method, the characteristics of training data are grouped, then corresponding RBF neural sub-networks are respectively trained, and finally, all RBF neural sub-networks are spliced in parallel, so that a network structure wider than a hidden layer of a traditional RBF neural network is obtained, the upper limit of the number of nodes of the hidden layer of the traditional RBF neural network is broken through, the RBF neural network can realize higher data classification capacity by means of more numbers of the nodes of the hidden layer and stronger data fitting and generalization capacity.
Example two
An embodiment of the present application provides a training apparatus 20 for an RBF neural network, and fig. 2 shows a schematic structural diagram of the training apparatus provided in the embodiment of the present application.
Specifically, referring to fig. 2, the training apparatus 20 includes an extracting module 21, a grouping module 22, a first training module 23, an integrating module 24, and a second training module 25.
The extraction module 21 is configured to extract features of each training data in the training set;
the grouping module 22 is configured to divide the features of each training data in the training set into Q feature subsets, where Q is greater than 1, and the number of features included in each feature subset is not less than 1;
the first training module 23 is configured to train corresponding RBF neural sub-networks based on each feature subset and a target training subset corresponding to each feature subset, respectively, to obtain Q RBF neural sub-networks, where training data in the target training subset includes and only includes all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same;
the integration module 24 is used for splicing the Q RBF neural sub-networks in parallel into an RBF neural network;
the second training module 25 is configured to train the RBF neural network based on the training data and the preset classification of the training data, so as to obtain a trained RBF neural network.
Optionally, the first training module 23 is specifically configured to calculate the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset; determining a target training subset corresponding to each characteristic subset by each characteristic subset; and respectively training the RBF neural sub-networks corresponding to the feature subsets based on the number of hidden layer nodes of the feature subsets and the target training subsets.
Optionally, the training device 20 further includes: a calculating module (not shown in the figure) for inputting each training data in the training set into the input layer of the RBF neural network, and calculating an output matrix of the hidden layer node of the RBF neural network; and calculating the weight from the hidden layer to the output layer in the RBF neural network based on the output matrix and the preset classification of each training data in the training set.
Optionally, the training device 20 further includes: an updating module (not shown in the figure) for updating the current parameters of the RBF neural network based on a gradient descent method, wherein the current parameters of the RBF neural network are formed by parameters of each RBF neural sub-network; the second training module 25 is specifically configured to: and training the updated RBF neural network based on the training data and the preset classification of the training data.
As can be seen from the above, the present application firstly extracts the features of each training data in the training set through the extraction module 21; dividing the features of each training data in the training set into Q feature subsets through a grouping module 22, wherein Q is greater than 1, and the number of features contained in each feature subset is not less than 1; then, training the corresponding RBF neural sub-networks based on each feature subset and the target training subset corresponding to each feature subset respectively by using a first training module 23 to obtain Q RBF neural sub-networks, wherein the training data in the target training subset includes and only includes all the features in the corresponding feature subset, and the input data and the output data of the RBF neural sub-networks are the same; then, the Q RBF neural sub-networks are spliced in parallel into an RBF neural network through the integration module 24; and finally, training the RBF neural network through a second training module 25 based on each training data and the preset classification of each training data to obtain the trained RBF neural network. According to the method, the characteristics of the training data are grouped, then the corresponding RBF neural sub-networks are respectively trained, and finally, all the RBF neural sub-networks are spliced in parallel, so that a network structure wider than a hidden layer of a traditional RBF neural network is obtained, the upper limit of the number of nodes of the hidden layer of the traditional RBF neural network is broken through, the RBF neural network can realize higher data classification capacity by using more nodes of the hidden layer and stronger data fitting and generalization capacity.
EXAMPLE III
Referring to fig. 3, the training apparatus includes a memory 31, a processor 32, and a computer program stored in the memory 31 and executable on the processor 32, wherein the memory 31 is used for storing software programs and modules, and the processor 32 executes the software programs and modules stored in the memory 31, so as to execute various functional applications and data processing. The memory 31 and the processor 32 are connected by a bus. In particular, the processor 32, by running the above-mentioned computer program stored in the memory 31, implements the following steps:
extracting the characteristics of each training data in the training set;
dividing the features of each training data in the training set into Q feature subsets, wherein Q is greater than 1, and the number of features contained in each feature subset is not less than 1;
training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subset comprises and only comprises all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same;
splicing the Q RBF neural sub-networks in parallel to form an RBF neural network;
and training the RBF neural network based on the training data and the preset classification of the training data to obtain the trained RBF neural network.
Assuming that the above is the first possible implementation manner, in a second possible implementation manner provided on the basis of the first possible implementation manner, the dividing the features of each piece of training data in the training set into Q feature subsets includes:
and dividing the characteristics of each training data in the training set into Q characteristic subsets according to a rule, wherein the rule is any one of a clustering division rule, a random division rule and a sequential division rule.
In a third possible implementation manner provided on the basis of the first possible implementation manner or the second possible implementation manner, the training the corresponding RBF neural sub-network based on each feature subset and the target training subset corresponding to each feature subset includes:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset;
determining a target training subset corresponding to each feature subset by each feature subset;
and respectively training the RBF neural sub-networks corresponding to the feature subsets based on the number of hidden layer nodes of the feature subsets and the target training subsets.
In a fourth possible implementation manner provided on the basis of the third possible implementation manner, the calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset specifically includes:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset based on a hidden layer node calculation formula, wherein the hidden layer node calculation formula specifically comprises the following steps:
wherein the content of the first and second substances,
is to take the integer operator, N'iIs the number of features contained in the ith feature subset, N is the number of features contained in the training set, and K is the number of target hidden layer nodes, K ', of the RBF neural network'iThe number of hidden layer nodes of the RBF neural sub-network.
In a fifth possible implementation form based on the first possible implementation form or the second possible implementation form, the training the RBF neural network based on the respective training data and the predetermined classification of the respective training data includes:
inputting each training data in the training set into an input layer of the RBF neural network, and calculating an output matrix of hidden layer nodes of the RBF neural network;
and calculating the weight from the hidden layer to the output layer in the RBF neural network based on the output matrix and the preset classification of each training data in the training set.
In a sixth possible implementation manner provided based on the first possible implementation manner or the second possible implementation manner, before the training the RBF neural network based on the respective training data and the preset classification of the respective training data, the method further includes:
updating the current parameters of the RBF neural network based on a gradient descent method, wherein the current parameters of the RBF neural network are formed by the parameters of each RBF neural sub-network;
the training of the RBF neural network based on the training data and the predetermined classification of the training data specifically includes:
and training the updated RBF neural network based on the training data and the preset classification of the training data.
It should be understood that, in the embodiment of the present Application, the Processor 32 may be a Central Processing Unit (CPU), and the Processor 32 may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components, and the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory 31 may include read-only memory, flash memory, and random access memory, and provides instructions and data to the processor. Some or all of the memory 31 may also include non-volatile random access memory.
As can be seen from the above, the present application first extracts the features of each training data in the training set; dividing the characteristics of each training data in the training set into Q characteristic subsets, wherein Q is greater than 1, and the number of the characteristics contained in each characteristic subset is not less than 1; training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subset comprises and only comprises all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same; splicing the Q RBF neural sub-networks in parallel to form an RBF neural network; and training the RBF neural network based on each training data and the preset classification of each training data to obtain the trained RBF neural network. According to the method, the characteristics of training data are grouped, then corresponding RBF neural sub-networks are respectively trained, and finally, all RBF neural sub-networks are spliced in parallel, so that a network structure wider than a hidden layer of a traditional RBF neural network is obtained, the upper limit of the number of nodes of the hidden layer of the traditional RBF neural network is broken through, the RBF neural network can realize higher data classification capacity by means of more numbers of the nodes of the hidden layer and stronger data fitting and generalization capacity.
It should be understood that the above-described integrated modules/units, if implemented in the form of software functional units and sold or used as separate products, may be stored in a computer readable storage medium. Based on such understanding, all or part of the flow in the method of the embodiments described above can be realized by a computer program, which can be stored in a computer-readable storage medium and can realize the steps of the embodiments of the methods described above when the computer program is executed by a processor. Wherein the computer program comprises computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the above-mentioned computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signal, telecommunication signal, software distribution medium, etc. It should be noted that the contents contained in the computer-readable storage medium can be increased or decreased as required by legislation and patent practice in the jurisdiction.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-mentioned functional units and modules are illustrated as being divided, and in practical applications, the above-mentioned functions may be distributed as different functional units and modules according to needs, that is, the internal structure of the apparatus may be divided into different functional units or modules to implement all or part of the above-mentioned functions. Each functional unit and module in the embodiments may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit, and the integrated unit may be implemented in the form of a hardware or a software functional unit. In addition, specific names of the functional units and modules are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present application. For the specific working processes of the units and modules in the system, reference may be made to the corresponding processes in the foregoing method embodiments, which are not described herein again.
It should be noted that, the methods and the details thereof provided by the foregoing embodiments may be combined with the apparatuses and devices provided by the embodiments, which are referred to each other and are not described again.
Those of ordinary skill in the art will appreciate that the elements and algorithm steps of the embodiments described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus/terminal device and method may be implemented in other ways. For example, the above-described apparatus/device embodiments are merely illustrative, and for example, the division of the above modules or units is only one logical functional division, and the actual implementation may be implemented by another division, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed.
The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.
Claims (10)
1. A training method based on an RBF neural network is characterized by comprising the following steps:
extracting the characteristics of each training data in the training set;
dividing the features of each training data in the training set into Q feature subsets, wherein Q is greater than 1, and the number of features contained in each feature subset is not less than 1;
training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subset comprises and only comprises all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same;
splicing the Q RBF neural sub-networks in parallel to form an RBF neural network;
and training the RBF neural network based on the training data and the preset classification of the training data to obtain the trained RBF neural network.
2. The training method of claim 1, wherein the dividing the features of the respective training data in the training set into Q feature subsets comprises:
and dividing the characteristics of each training data in the training set into Q characteristic subsets according to rules, wherein the rules are any one of clustering division rules, random division rules and sequence division rules.
3. The training method according to claim 1 or 2, wherein the training of the corresponding RBF neural sub-network based on the respective feature subsets and the target training subsets corresponding to the respective feature subsets, respectively, comprises:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset;
determining a target training subset corresponding to each feature subset by each feature subset;
and respectively training the RBF neural sub-networks corresponding to the feature subsets based on the number of hidden layer nodes of the feature subsets and the target training subsets.
4. The training method according to claim 3, wherein the calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset is specifically:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset based on a hidden layer node calculation formula, wherein the hidden layer node calculation formula specifically comprises the following steps:
wherein the content of the first and second substances,
is to take the integer operator, N'iThe number of features included in the ith feature subset, N being the number of features included in the training setThe number of contained features, K, is the number of target hidden layer nodes, K ', of the RBF neural network'iIs the number of hidden layer nodes of the RBF neural sub-network.
5. The training method of claim 1 or 2, wherein the training the RBF neural network based on the respective training data and the preset classification of the respective training data comprises:
inputting each training data in the training set into an input layer of the RBF neural network, and calculating an output matrix of nodes of a hidden layer of the RBF neural network;
and calculating the weight from a hidden layer to an output layer in the RBF neural network according to the output matrix and the preset classification of each training data in the training set.
6. The training method of claim 1 or 2, wherein said training the RBF neural network based on the respective training data and the predetermined classification of the respective training data further comprises:
updating the current parameters of the RBF neural network based on a gradient descent method, wherein the current parameters of the RBF neural network are composed of the parameters of each RBF neural sub-network;
the training of the RBF neural network based on the training data and the preset classification of the training data specifically includes:
and training the updated RBF neural network based on the training data and the preset classification of the training data.
7. An RBF neural network-based training device, comprising:
the extraction module is used for extracting the characteristics of each training data in the training set;
the grouping module is used for dividing the characteristics of each training data in the training set into Q characteristic subsets, wherein Q is greater than 1, and the number of the characteristics contained in each characteristic subset is not less than 1;
the first training module is used for training corresponding RBF neural sub-networks respectively based on each feature subset and a target training subset corresponding to each feature subset to obtain Q RBF neural sub-networks, wherein training data in the target training subset comprises and only comprises all features in the corresponding feature subset, and input data and output data of the RBF neural sub-networks are the same;
the integration module is used for splicing the Q RBF neural sub-networks into an RBF neural network in parallel;
and the second training module is used for training the RBF neural network based on the training data and the preset classification of the training data to obtain the trained RBF neural network.
8. The training apparatus of claim 7, wherein the first training module is specifically configured to:
respectively calculating the number of hidden layer nodes of the RBF neural sub-network corresponding to each feature subset;
determining a target training subset corresponding to each feature subset by each feature subset;
and respectively training the RBF neural sub-networks corresponding to the feature subsets based on the number of hidden layer nodes of the feature subsets and the target training subsets.
9. An RBF neural network-based training device, comprising: memory, processor and computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 6 when executing the computer program.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911309763.3A CN111079843A (en) | 2019-12-18 | 2019-12-18 | Training method based on RBF neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911309763.3A CN111079843A (en) | 2019-12-18 | 2019-12-18 | Training method based on RBF neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111079843A true CN111079843A (en) | 2020-04-28 |
Family
ID=70315472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911309763.3A Pending CN111079843A (en) | 2019-12-18 | 2019-12-18 | Training method based on RBF neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111079843A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733995A (en) * | 2021-01-07 | 2021-04-30 | 中国工商银行股份有限公司 | Method for training neural network, behavior detection method and behavior detection device |
-
2019
- 2019-12-18 CN CN201911309763.3A patent/CN111079843A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733995A (en) * | 2021-01-07 | 2021-04-30 | 中国工商银行股份有限公司 | Method for training neural network, behavior detection method and behavior detection device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11651259B2 (en) | Neural architecture search for convolutional neural networks | |
| US11829874B2 (en) | Neural architecture search | |
| US20190087713A1 (en) | Compression of sparse deep convolutional network weights | |
| CN109034371B (en) | Deep learning model reasoning period acceleration method, device and system | |
| CN112365171B (en) | Knowledge graph-based risk prediction method, device, equipment and storage medium | |
| EP3686813A1 (en) | Deep neural network-based method and device for quantifying activation amount | |
| WO2016182659A1 (en) | Bit width selection for fixed point neural networks | |
| CN104346629A (en) | Model parameter training method, device and system | |
| JP2022547460A (en) | Performing XNOR equivalent operations by adjusting column thresholds of compute-in-memory arrays | |
| CN111563192A (en) | Entity alignment method and device, electronic equipment and storage medium | |
| CN112464233B (en) | RNN-based malicious software detection method on cloud platform | |
| CN110705585A (en) | Network fraud identification method and device, computer device and storage medium | |
| CN108805174A (en) | clustering method and device | |
| CN113887704A (en) | Traffic information prediction method, device, equipment and storage medium | |
| CN113743650B (en) | Power load prediction method, device, equipment and storage medium | |
| CN110874635B (en) | Deep neural network model compression method and device | |
| CN111079843A (en) | Training method based on RBF neural network | |
| CN113168554B (en) | Neural network compression method and device | |
| CN108681490B (en) | Vector processing method, device and equipment for RPC information | |
| CN111275166A (en) | Image processing device and equipment based on convolutional neural network and readable storage medium | |
| CN112800419A (en) | Method, apparatus, medium and device for identifying IP group | |
| CN114496068A (en) | Protein secondary structure prediction method, device, equipment and storage medium | |
| CN116258190A (en) | Quantization method, quantization device and related equipment | |
| CN112489687A (en) | Speech emotion recognition method and device based on sequence convolution | |
| WO2021055364A1 (en) | Efficient inferencing with fast pointwise convolution |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |