CN115422989A - Unsupervised clustering algorithm and unsupervised clustering device for neural network and electronic equipment thereof - Google Patents
Unsupervised clustering algorithm and unsupervised clustering device for neural network and electronic equipment thereof Download PDFInfo
- Publication number
- CN115422989A CN115422989A CN202210875405.4A CN202210875405A CN115422989A CN 115422989 A CN115422989 A CN 115422989A CN 202210875405 A CN202210875405 A CN 202210875405A CN 115422989 A CN115422989 A CN 115422989A
- Authority
- CN
- China
- Prior art keywords
- network
- reference network
- neural
- layer
- matching
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- 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
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an unsupervised clustering algorithm, a device and electronic equipment thereof for a neural network, which relate to the technical field of networks and comprise the following steps: acquiring a plurality of neural networks, and selecting a reference network from the neural networks; acquiring the similarity of implicit elements of a reference network and a neural network except the reference network; performing maximum dichotomy weight matching layer by layer based on the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result; and according to the matching result, adjusting the weight connection sequence of each layer of the neural network except the reference network. The clustering accuracy and the performance of the aggregated model are greatly improved.
Description
Technical Field
The invention belongs to the technical field of networks, and particularly relates to an unsupervised clustering algorithm and device for a neural network.
Background
With the continuous development of the internet of things, mobile phones, wearable devices, autonomous vehicles and the like all use distributed networks; among other things, devices in a distributed network can generate large amounts of data and create a need for multiple task processing, challenging the computing power of the devices.
In the prior art, collaborative learning participated by a plurality of clients is promoted, so that the computing capability of the equipment is continuously improved, and the equipment can locally store data and complete computation; the collaborative learning is widely applied as a learning framework which can be used for large-scale training on equipment for generating data; however, devices in the distributed network generate a large amount of data, and processing the local large amount of data in time through the neural network is an urgent problem to be solved at present, and the association between the clients cannot be effectively utilized by using collaborative learning, so that higher clustering accuracy cannot be obtained.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an unsupervised clustering algorithm and a device thereof facing to a neural network. The technical problem to be solved by the invention is realized by the following technical scheme:
in a first aspect, the present application provides an unsupervised clustering algorithm for a neural network, including:
acquiring a plurality of neural networks, and selecting a reference network from the neural networks;
acquiring the similarity of implicit elements of a reference network and a neural network except the reference network;
carrying out maximum dichotomy weight matching layer by layer based on the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result;
and according to the matching result, adjusting the weight connection sequence of each layer of the neural network except the reference network.
Optionally, a plurality of neural networks are obtained, and a reference network is selected from the neural networks, which includes two modes: selecting through model performance and selecting through discrimination;
the process of selecting through model performance includes: classifying the same test data by adopting different neural networks respectively, and taking the neural network with the highest classification accuracy as a reference network;
the process of selecting through discrimination comprises the following steps: and respectively obtaining the variances of the plurality of neural networks, and selecting a reference network according to the variances of the neural networks.
Optionally, the maximum binary weight matching is performed layer by layer based on the implicit element similarity of the reference network and the implicit element similarity of the neural network except for the reference network, and the process of obtaining the matching result includes:
selecting a reference network and a corresponding layer of a neural network except the reference network to construct a plurality of bipartite graphs;
acquiring the maximum matching of all constructed bipartite graphs by using a Hungarian algorithm;
based on the maximum matching of all constructed bipartite graphs, acquiring the maximum weight matching of all constructed bipartite graphs by using a KM algorithm;
and acquiring corresponding layers of the reference network and the neural network except the reference network layer by layer, constructing a plurality of bipartite graphs in each layer, and acquiring maximum weight matching of all constructed bipartite graphs to obtain a matching result.
Optionally, the method for obtaining the hidden element similarity of the reference network includes euclidean distance, hamming distance, or cosine similarity.
In a second aspect, the present application further provides an unsupervised clustering apparatus for a neural network, including:
the selection module is used for acquiring a plurality of neural networks and selecting a reference network from the neural networks;
the acquisition module is used for acquiring the similarity of implicit elements of the reference network and the neural network except the reference network;
the matching module is used for carrying out maximum dichotomy weight matching layer by layer based on the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result;
and the weight connection module is used for adjusting the weight connection sequence of each layer of the neural network except the reference network according to the matching result.
In a third aspect, the present application further provides an electronic device, including: a processor and a machine-readable storage medium;
a machine-readable storage medium stores machine-executable instructions executable by a processor;
the processor is configured to execute machine-executable instructions to implement the algorithm steps in the above-described embodiments.
The invention has the beneficial effects that:
the unsupervised clustering algorithm, the unsupervised clustering device and the electronic equipment thereof for the neural network, provided by the invention, have the advantages that the reference network is obtained from a plurality of neural networks, the similarity of hidden elements of each layer of the neural networks is obtained, the maximum dichotomy weight matching is carried out layer by layer, and the influence of the displacement invariance of the hidden elements in the neural networks is eliminated; in addition, for different networks in distributed learning, the clustering accuracy and the performance of the aggregated model are greatly improved by realizing the matching of hidden elements of each layer.
The present invention will be described in further detail with reference to the drawings and examples.
Drawings
FIG. 1 is a flowchart of an unsupervised clustering algorithm for neural networks according to an embodiment of the present invention;
FIG. 2 is a bipartite graph according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an unsupervised clustering apparatus for a neural network according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
In the prior art, clustering is a process of dividing a set of physical or abstract objects into a plurality of classes composed of similar objects, and a cluster generated by clustering is a set of data objects, which are similar to objects in the same cluster and different from objects in other clusters. Because the neural network is uploaded by the client side instead of the original training data, how to cluster the ultrahigh-dimensional data with millions or millions of parameters becomes a difficult point. In addition, the relation of implicit elements in the neural network (such as convolution kernels in the convolution neural network and hidden states in the long-short term memory network) is parallel, that is, the extraction of the same feature may be responsible for elements at different positions on different networks, which also hinders the implementation of the clustering or aggregation method.
In view of this, the present application provides an unsupervised clustering algorithm and a device thereof for a neural network, which are designed for the neural network in distributed learning, so as to better mine the internal connection between networks.
Referring to fig. 1, fig. 1 is a flowchart of an unsupervised clustering algorithm for a neural network according to an embodiment of the present invention, where the unsupervised clustering algorithm for a neural network provided in the present application includes:
s101, obtaining a plurality of neural networks and selecting a reference network from the neural networks;
s102, obtaining the similarity of implicit elements of a reference network and a neural network except the reference network;
s103, performing maximum dichotomy weight matching layer by layer based on the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result;
and S104, adjusting the weight connection sequence of each layer of the neural network except the reference network according to the matching result.
Specifically, the unsupervised clustering algorithm for the neural network provided in this embodiment is designed for the defects that the neural network parameters uploaded by the client in the distributed network are many, the dimensionality is high, and hidden elements in different neural networks may adversely affect the extraction of the same feature, so that the internal connection among networks can be better mined, the replacement invariance of the hidden elements in the neural network is eliminated, and the clustering accuracy and the model performance are improved.
In an optional embodiment of the present application, the obtaining of the plurality of neural networks and the selecting of the reference network therefrom include two ways, which are respectively: selecting through model performance and selecting through discrimination;
the process of selecting by model performance includes: classifying the same test data by adopting different neural networks respectively, and taking the neural network with the highest classification accuracy as a reference network;
the process of selecting through discrimination comprises the following steps: the variances of the plurality of neural networks are respectively obtained, and the neural network with the proper variance is selected as a reference network.
In an optional embodiment of the present application, please refer to fig. 2, where fig. 2 is a bipartite graph provided in an embodiment of the present invention, and based on the implicit element similarity of the reference network and the implicit element similarity of the neural network except for the reference network, the process of performing maximum dichotomy weight matching layer by layer to obtain a matching result includes:
selecting corresponding layers of a reference network and a neural network except the reference network to construct a plurality of bipartite graphs;
obtaining the maximum matching of all constructed bipartite graphs by using a Hungarian algorithm;
based on the maximum matching of all constructed bipartite graphs, acquiring the maximum weight matching of all constructed bipartite graphs by using a KM algorithm;
and acquiring corresponding layers of the reference network and the neural network except the reference network layer by layer, constructing a plurality of bipartite graphs in each layer, and acquiring maximum weight matching of all constructed bipartite graphs to obtain a matching result.
In fig. 2, the other networks are neural networks except the reference network, where a, B, and C are the number of layers of the reference network, and a, B, and C are the other neural networks.
In an optional embodiment of the present application, the method for obtaining the similarity of the hidden elements of the reference network includes euclidean distance, hamming distance, or cosine similarity.
Referring to fig. 1 and fig. 2, in an optional embodiment of the present application, an unsupervised clustering algorithm for a neural network includes:
s101, building a selection frame of a reference network, and selecting the reference network from a plurality of neural networks uploaded by a client through the selection frame of the reference network.
Optionally, the neural network uploaded by the client is generally a Convolutional Neural Network (CNN) or a long short term memory network (LSTM); the CNN is composed of a convolutional layer, a pooling layer, a full-link layer and the like, and is used for extracting the characteristics of a certain object according to a certain model and classifying, identifying, deciding or predicting the object according to the extracted characteristics; the LSTM is a feedback neural network (RNN) mainly used for processing time series data and widely applied to scenes such as stock trend prediction, relation classification in NLP (non line segment prediction) and speech recognition.
In the embodiment, the reference network is selected according to the requirement of the distributed learning task, and the selection standard is divided into two types according to different requirements, namely, the selection through model performance and the selection through discrimination.
The method comprises the following steps that a reference network is selected according to model performance, and the model performance is reflected by the accuracy of a neural network to complete a distributed learning task; for example, a plurality of clients all use the CNN to perform image classification tasks, a reference network is selected by comparing the accuracy of classification of the same test data of each client model, and a neural network with the highest accuracy is selected as the reference network of the unsupervised neural network clustering algorithm.
Selecting a reference network through discrimination, wherein the discrimination is embodied by the variance of the neural network, and the higher the variance is in a certain range, the higher the discrimination of the network is; for example, a complex classifier with a very deep depth and many hidden nodes is used for fitting a data set, a large and deep neural network can fully learn the characteristics of a sample, but if the setting is not good, an excessively high variance is obtained, so that misclassification occurs; a more appropriate variance can be obtained by using methods such as regularization or increasing the number of samples; it can be understood that when there is a strict requirement on the neural network classification standard, a network with a proper variance needs to be selected as a reference network.
In the embodiment, a reference network is selected from a plurality of neural networks uploaded by a client, so that the characteristic matching is carried out to eliminate the permutation invariance of hidden elements of the neural networks; the selection mode of the reference network is not limited in this embodiment, and those skilled in the art can select the reference network according to the requirement of the distributed learning task.
S102, calculating the similarity of the implicit elements of the reference network and the neural network except the reference network.
Optionally, the implicit elements refer to different types according to different uploading networks of users, for example, a convolution kernel in a convolutional neural network and a hidden state in a long-term and short-term memory network, and the relationship of the implicit elements in the neural network is parallel, that is, the extraction of the same feature may be responsible for elements at different positions on different networks, so that the implementation of a clustering or aggregation method may be hindered.
In the embodiment, by calculating the similarity of the hidden elements of the reference network and the neural network except the reference network, the hidden elements with higher similarity represent higher probability of being responsible for the same feature extraction, and the calculated similarity is used for performing binary maximum weight matching to provide a weight value reference; alternatively, the method for calculating the similarity includes, but is not limited to, euclidean distance, hamming distance, cosine similarity, or the like.
S103, carrying out maximum dichotomy weight matching layer by layer based on the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result.
And S1031, selecting the reference network and the corresponding layer of the neural network except the reference network, and constructing a plurality of bipartite graphs.
The reference network is one of n neural networks uploaded by a client selected through a reference network selection frame, n-1 bipartite graphs are constructed according to the corresponding relation between the implicit elements and the extracted features of the corresponding layers of the reference network and the neural networks except the reference network, the implicit elements of the corresponding layers of the reference network and the neural networks except the reference network are selected to construct the bipartite graphs, wherein the connecting line between the implicit elements is determined according to the similarity between the implicit elements acquired in the step S102, the threshold value of the similarity in the embodiment is set to be 0.6, and if the similarity between the two implicit elements is greater than or equal to 0.6, the extracted features of the two implicit elements are consistent or related; if the similarity between the two hidden elements is less than 0.6, extracting the characteristics of the two hidden elements is inconsistent, and the two hidden elements cannot be connected; the threshold of the connection line between the bipartite graphs is set by a person skilled in the art according to service needs, and the embodiment is not limited herein.
S1032, obtaining the maximum matching of all constructed bipartite graphs by using the Hungarian algorithm.
A match is a set of edges (i.e., the links in the above step) where no two edges have a common vertex; the maximum matching refers to the matching with the largest number of matching edges in all the matches, wherein the maximum matching relates to two concepts, namely an alternate path and an augmented path, starting from a matching point, and sequentially passing through paths formed by non-matching edges, non-matching edges and the like, namely the alternate path; starting from one unmatched point, an alternate path is taken, and if the path is routed to another unmatched point (non-starting point), the alternate path is an augmented path.
In this embodiment, by using the hungarian algorithm, the match with the largest number of matching edges is finally found through repeated iterations by continuously finding the augmented path of the constructed bipartite graph, and the obtained maximum matching result is a precondition for obtaining the maximum bipartite matching by performing the KM algorithm.
And S1033, based on the maximum matching of all constructed bipartite graphs, acquiring the maximum weight matching of all constructed bipartite graphs by using a KM algorithm.
In order to facilitate processing by using a KM algorithm, the similarity value of the privacy elements obtained in the step S102 is expanded; the expansion factor is determined according to actual conditions, and the application is not limited herein.
In this embodiment, based on the maximum matching obtained in step S1032, a KM algorithm is used to add an edge weight to the bipartite graph, and the maximum weight matching is obtained by calculating the maximum value of the matching edge weight sum; setting the point weight of the left point as the maximum side weight, setting the right point side weight as 0, and matching when the side weights of the left point and the right point and the side weight equal to the side where the two points are connected; and matching is carried out again when the conflict occurs and the last point is withdrawn to have no side which can be matched, the left side point side right is minus 1, the right side point side right is plus 1, and if the match cannot be carried out, the left side point side right is minus 1 and the right side point side right is plus 1 continuously until the match is successful.
S1034, repeating the steps S1031 to S1033 layer by layer until the matching of the reference network and the final corresponding layer of the neural network except the reference network is completed, and obtaining a matching result.
And finding the maximum weight matching by carrying out the KM algorithm layer by layer, adjusting the overall structure of the neural network except the reference network, and realizing the overall matching of the neural network except the reference network and the reference network.
In the embodiment, the similarity is calculated by calculating hidden elements in the neural network, the KM algorithm is adopted for layer-by-layer maximum dichotomy matching, the neural network structures except the reference network are adjusted to be matched with the reference network, namely other structures are adjusted to meet the requirements of the distributed learning task, and therefore the clustering accuracy and the performance of the aggregated model can be greatly improved.
And S104, adjusting the weight connection sequence of each layer of the neural network except the reference network according to the matching result.
And according to the result of the maximum weight matching, adjusting the weight connection sequence of the neural networks except the reference network, eliminating the influence of the invariance of the replacement of hidden elements in the neural networks, and greatly improving the clustering accuracy and the performance of the aggregated model.
Through the steps, in the scene of distributed learning, the neural networks uploaded by the client are obtained by establishing a set of complete unsupervised neural network clustering algorithm rules, the reference networks are selected according to needs through the reference network selection framework, the similarity of hidden elements of the neural networks is calculated, layer-by-layer maximum binary weight matching is performed through the KM algorithm, the influence of invariance of replacement of the hidden elements in the neural networks is eliminated, and therefore clustering accuracy and aggregated model performance are greatly improved.
Based on the same inventive concept, please refer to fig. 3, where fig. 3 is a schematic structural diagram of an unsupervised clustering device for a neural network provided in an embodiment of the present invention, and the present application also provides an unsupervised clustering device for a neural network, which is applied to an unsupervised clustering algorithm for a neural network provided in the above embodiment of the present application, and the algorithm may refer to the above embodiment and is not described herein again; the device comprises:
a selecting module 201, configured to obtain a plurality of neural networks and select a reference network from the neural networks;
an obtaining module 202, configured to obtain an implicit element similarity between a reference network and a neural network other than the reference network;
the matching module 203 is used for performing maximum dichotomy weight matching layer by layer based on the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result;
and a weight connection module 204, configured to adjust a weight connection order of each layer of the neural network except the reference network according to the matching result.
Specifically, in the unsupervised clustering device for the neural network provided by this embodiment, for increasingly-used distributed learning scenes, a reference network is selected according to task requirements, similarity of hidden elements of the model is obtained, and a KM algorithm is adopted to perform maximum binary weight distribution layer by layer, so that influence of invariance of replacement of the hidden elements in the neural network is greatly eliminated, and clustering accuracy and aggregated model performance can be greatly improved through the unsupervised neural network clustering algorithm.
Based on the same inventive concept, please refer to fig. 4, fig. 4 is a schematic structural diagram of an electronic device provided in an embodiment of the present invention, and the present application further provides an electronic device, including: a processor 301 and a machine-readable storage medium 303;
the machine-readable storage medium 303 stores machine-executable instructions executable by the processor 301;
the processor 301 is configured to execute machine executable instructions to implement an unsupervised clustering algorithm for a neural network provided by the above-described embodiments.
Specifically, as shown in fig. 4, the electronic device provided in this embodiment further includes a communication interface 302 and a communication bus 304, where the processor 301, the communication interface 302, and the machine-readable storage medium 303 realize communication with each other through the communication bus 304. The processor 301 may include one or more Processing cores, such as a 4-core processor, an 8-core processor, etc., and the processor 301 may be implemented in at least one hardware form of Digital Signal Processing (DSP), field-Programmable Gate Array (FPGA), programmable Logic Array (PLA); the processor 301 may also include a main processor and a coprocessor, where the main processor is a processor for Processing data in a wake state, and is also called a Central Processing Unit (CPU); a coprocessor is a low power processor for processing data in a standby state. Alternatively, the processor 301 may be a Graphics Processing Unit (GPU) integrated with the GPU, and the GPU is responsible for rendering and drawing the content required to be displayed by the display screen.
The machine-readable storage medium 303 may include one or more computer-readable storage media, which may be non-transitory. The machine-readable storage medium 303 may also include high-speed random access memory, as well as non-volatile memory, such as one or more magnetic disk storage devices, flash memory storage devices. Optionally, the machine-readable storage medium 303 is at least used for storing a computer program, wherein the computer program can implement the unsupervised neural network clustering algorithm provided by the above-mentioned embodiments after being loaded and executed by the processor. In addition, the resources stored by the machine-readable storage medium 303 may also include an operating system, data, and the like, and the storage manner may be transient storage or permanent storage; the operating system may include Windows, unix, linux, etc.; the data can include, but is not limited to, data involved in network behavior analysis and situation awareness methods oriented to dynamic data, and the like.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Furthermore, these computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The unsupervised clustering algorithm, the unsupervised clustering device and the electronic equipment thereof for the neural network, provided by the invention, have the advantages that the reference network is obtained from a plurality of neural networks, the similarity of hidden elements of each layer of the neural networks is obtained, the maximum dichotomy weight matching is carried out layer by layer, and the influence of the displacement invariance of the hidden elements in the neural networks is eliminated; in addition, for different networks in distributed learning, the clustering accuracy and the performance of the aggregated model are greatly improved by realizing the matching of hidden elements of each layer.
The foregoing is a further detailed description of the invention in connection with specific preferred embodiments and it is not intended to limit the invention to the specific embodiments described. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (6)
1. An unsupervised clustering algorithm for neural networks, comprising:
acquiring a plurality of neural networks, and selecting a reference network from the neural networks;
obtaining the similarity of the implicit elements of the reference network and the neural network except the reference network;
performing maximum dichotomy weight matching layer by layer based on the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result;
and adjusting the weight connection sequence of each layer of the neural network except the reference network according to the matching result.
2. The unsupervised clustering algorithm based on neural networks as claimed in claim 1, wherein the obtaining of a plurality of neural networks and the selecting of the reference network therefrom comprise two modes, respectively: selecting through model performance and selecting through discrimination;
the process of selecting through model performance includes: classifying the same test data by adopting different neural networks respectively, and taking the neural network with the highest classification accuracy as a reference network;
the process of selecting through discrimination comprises the following steps: and respectively obtaining the variances of the plurality of neural networks, and selecting a reference network according to the variances of the neural networks.
3. The unsupervised clustering algorithm based on the neural network as claimed in claim 1, wherein the maximum dichotomy weight matching is performed layer by layer based on the implicit element similarity of the reference network and the implicit element similarity of the neural network except the reference network, and the process of obtaining the matching result comprises:
selecting the reference network and the corresponding layers of the neural networks except the reference network to construct a plurality of bipartite graphs;
obtaining the maximum matching of all constructed bipartite graphs by using a Hungarian algorithm;
based on the maximum matching of all constructed bipartite graphs, acquiring the maximum weight matching of all constructed bipartite graphs by using a KM algorithm;
and acquiring corresponding layers of the reference network and the neural network except the reference network layer by layer, constructing a plurality of bipartite graphs for each layer, and acquiring the maximum weight matching of all constructed bipartite graphs to obtain a matching result.
4. The unsupervised clustering algorithm based on neural network as claimed in claim 1, wherein the method for obtaining the similarity of the implicit elements of the reference network comprises Euclidean distance, hamming distance or cosine similarity.
5. An unsupervised clustering device oriented to a neural network, comprising:
the selection module is used for acquiring a plurality of neural networks and selecting a reference network from the neural networks;
the acquisition module is used for acquiring the similarity of the implicit elements of the reference network and the neural network except the reference network;
the matching module is used for carrying out maximum dichotomy weight matching layer by layer on the basis of the hidden element similarity of the reference network and the hidden element similarity of the neural network except the reference network to obtain a matching result;
and the weight connection module is used for adjusting the weight connection sequence of each layer of the neural network except the reference network according to the matching result.
6. An electronic device, comprising: a processor and a machine-readable storage medium;
the machine-readable storage medium stores machine-executable instructions executable by the processor;
the processor is configured to execute machine executable instructions to implement the algorithm steps of any of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210875405.4A CN115422989A (en) | 2022-07-21 | 2022-07-21 | Unsupervised clustering algorithm and unsupervised clustering device for neural network and electronic equipment thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210875405.4A CN115422989A (en) | 2022-07-21 | 2022-07-21 | Unsupervised clustering algorithm and unsupervised clustering device for neural network and electronic equipment thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115422989A true CN115422989A (en) | 2022-12-02 |
Family
ID=84196712
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210875405.4A Pending CN115422989A (en) | 2022-07-21 | 2022-07-21 | Unsupervised clustering algorithm and unsupervised clustering device for neural network and electronic equipment thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115422989A (en) |
-
2022
- 2022-07-21 CN CN202210875405.4A patent/CN115422989A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210334624A1 (en) | Neural architecture search using a performance prediction neural network | |
| US11669744B2 (en) | Regularized neural network architecture search | |
| US11741361B2 (en) | Machine learning-based network model building method and apparatus | |
| US11636314B2 (en) | Training neural networks using a clustering loss | |
| CN109948149B (en) | Text classification method and device | |
| WO2023138188A1 (en) | Feature fusion model training method and apparatus, sample retrieval method and apparatus, and computer device | |
| US9875294B2 (en) | Method and apparatus for classifying object based on social networking service, and storage medium | |
| WO2022042123A1 (en) | Image recognition model generation method and apparatus, computer device and storage medium | |
| EP3620982B1 (en) | Sample processing method and device | |
| EP3839764A1 (en) | Method and system for detecting duplicate document using vector quantization | |
| US20230298314A1 (en) | Image clustering method and apparatus, computer device, and storage medium | |
| US20220343146A1 (en) | Method and system for temporal graph neural network acceleration | |
| CN112418320B (en) | Enterprise association relation identification method, device and storage medium | |
| CN110489955B (en) | Image processing, device, computing device and medium applied to electronic equipment | |
| US20220004849A1 (en) | Image processing neural networks with dynamic filter activation | |
| CN111882048A (en) | Neural network structure searching method and related equipment | |
| US11734573B2 (en) | Image element matching via graph processing | |
| CN116127386B (en) | Sample classification method, device, equipment and computer readable storage medium | |
| CN115422989A (en) | Unsupervised clustering algorithm and unsupervised clustering device for neural network and electronic equipment thereof | |
| CN111091198A (en) | Data processing method and device | |
| CN114638823B (en) | Full-slice image classification method and device based on attention mechanism sequence model | |
| CN112445933B (en) | Model training method, device, equipment and storage medium | |
| CN116957006A (en) | Training method, device, equipment, medium and program product of prediction model | |
| CN114359811A (en) | Data authentication method and device, electronic equipment and storage medium | |
| CN113723593A (en) | Load shedding prediction method and system based on neural network |
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 |