CN113591913B - Picture classification method and device supporting incremental learning - Google Patents
Picture classification method and device supporting incremental learning Download PDFInfo
- Publication number
- CN113591913B CN113591913B CN202110716563.0A CN202110716563A CN113591913B CN 113591913 B CN113591913 B CN 113591913B CN 202110716563 A CN202110716563 A CN 202110716563A CN 113591913 B CN113591913 B CN 113591913B
- Authority
- CN
- China
- Prior art keywords
- picture
- convolutional neural
- training
- neural network
- classifier
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 41
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 27
- 239000013598 vector Substances 0.000 claims abstract description 26
- 238000012360 testing method Methods 0.000 claims abstract description 20
- 238000013434 data augmentation Methods 0.000 claims abstract description 4
- 230000001537 neural effect Effects 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 14
- 230000009977 dual effect Effects 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000003213 activating effect Effects 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 210000005036 nerve Anatomy 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000007789 sealing Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a picture classification method and device supporting incremental learning, wherein the method comprises the following steps: selecting training picture samples in a training set, inputting the training picture samples into a convolutional neural network, and selecting typical picture samples through the convolutional neural network; updating the representative memory by a typical picture sample; performing data augmentation on the data set of the representative memory and the increment task to construct a test set; inputting the test picture samples in the test set into a convolutional neural network, and extracting feature vectors through the convolutional neural; and inputting the feature vectors of all the test picture samples into a trained classifier, and outputting a picture classification result through the increment of the classifier. The invention can learn the picture characteristics and classify the pictures on line, thereby improving the classification accuracy.
Description
Technical Field
The invention relates to a picture classification method and device supporting incremental learning, and belongs to the technical field of image classification.
Background
Most of traditional machine learning modes are off-line learning and batch learning, the off-line learning and batch learning modes can not dynamically learn data characteristics, and meanwhile, a large amount of memory space is required to store historical data. Incremental learning aims to develop an algorithm that does not stop learning, but rather continuously updates model parameters over time. In recent years, with the remarkable increase of large data, incremental learning has become a popular direction.
In recent years, many research results have been presented in this field. Early incremental learning algorithms were based in large part on feed forward neural networks. ParekhR et al teach a method for incremental learning by adjusting the structure of a multi-layer perceptron neural network (MLP). When the knowledge contained in the new sample is not in the learned knowledge range of the MLP network, incremental learning of the new sample is realized by adding hidden layer units. However, when the new sample size is large, difficulty will occur in implementing incremental learning by means of weight adjustment and structure adjustment. The passive active algorithm (passive aggressive learning, PA) is a second order algorithm that uses this information to assist model update in learning taking into account the confidence level of the prediction, i.e. the sample-to-current decision boundary spacing, and the PA uses more complex learning steps: the learning step size on each sample is related to the confidence level that the sample is classified. Confidence-weighted learning (CW) is another second order algorithm that is an extension of the PA algorithm. The algorithm CW always forces the current sample to be correctly classified, which constraint makes it very susceptible to noise data to produce an overfitting.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provides a picture classification method and device supporting incremental learning, which can be used for classifying pictures on line and remarkably improving the classification accuracy.
In order to achieve the above purpose, the invention is realized by adopting the following technical scheme:
in a first aspect, the present invention provides a method for classifying pictures, including:
selecting training picture samples in a training set, inputting the training picture samples into a convolutional neural network, and selecting typical picture samples through the convolutional neural network;
updating the representative memory by a typical picture sample;
performing data augmentation on the data set of the representative memory and the increment task to construct a test set;
inputting the test picture samples in the test set into a convolutional neural network, and extracting feature vectors through the convolutional neural;
and inputting the feature vectors of all the test picture samples into a trained classifier, and outputting a picture classification result through the increment of the classifier.
Preferably, the selecting the typical picture sample through the convolutional neural network includes: the convolutional neural network outputs posterior probability according to training picture samples which can be correctly identified in the input training picture samples, and typical picture samples are selected from the training picture samples according to the sequence of the posterior probability from high to low.
Preferably, the convolutional neural network includes: 8 convolution layers, wherein each convolution layer uses a convolution kernel of 3×3 and 64 nodes, maxPooling layers are added after the 2 nd, 4 th, 6 th and 8 th convolution layers, the pool size of each MaxPooling layer is 2×2, and the step length is 2; and adding a Dropout layer on the last MaxPooling layer, wherein the Dropout probability is 0.5.
Preferably, the extracting the feature vector by convolution nerve includes:
carrying out convolution processing on an input test picture sample through a convolution layer to obtain image characteristics;
normalizing the image features through batch normalization;
carrying out pooling treatment on the normalized image features through a MaxPooling layer;
and activating the pooled image features through a ReLU activation function to obtain feature vectors.
Preferably, the classifier comprises a regular dual average algorithm RDA and a soft confidence weighting algorithm SCW, wherein the regular dual average algorithm RDA is used for improving the feature extraction capability of the stream data, and the soft confidence weighting algorithm SCW is used for outputting classification results.
Preferably, the expression of the optimization problem of the soft confidence weighting algorithm SCW is as follows:
where N (μ, Σ) represents the Gaussian distribution of the classifier obeying the mean vector μ and covariance matrix Σ, D KL Representing a new Gaussian distribution N (μ) t+1 ,Σ t+1 ) And the current Gaussian distribution N (mu) t ,Σ t ) Degree of divergence, iota between Φ As a loss function, C represents a trade-off between passive and aggressive parameters, (x) t ,y t ) Representing the input feature vector;
the sealing solution is as follows:
wherein alpha is t And beta t The coefficients, respectively, are expressed as follows:
wherein,m t =y t (μ t ·x t );Φ=Φ -1 (eta); phi represents a cumulative function of normal distribution, phi -1 (eta) is an inverse function of phi, which is used to generate random variables subject to random distribution, eta e [0,1]。
Preferably, the expression of the weight update strategy of the regular dual average algorithm RDA is as follows:
wherein,<G (t) ,W>representing gradient G (t) The integrated average value of the weight W, ψ (W) is a regular term, ψ (W) = lambda W 1 Lambda denotes the learning rate, h (W) is an auxiliary rigid convex function,β (t) is a non-negative and non-self-decreasing sequence, < >>t represents a time step.
Preferably, the training of the classifier includes:
selecting training picture samples in a training set, inputting the training picture samples into a convolutional neural network, and extracting feature vectors through the convolutional neural network;
the classifier outputs a prediction label according to the feature vector of the input training picture sample;
judging whether the predicted label is consistent with the actual label of the training picture sample,
if yes, the classifier is kept unchanged, and if not, parameters of the classifier are updated.
In a second aspect, the present invention provides a picture classification apparatus supporting incremental learning, including a processor and a storage medium;
the storage medium is used for storing instructions;
the processor is operative according to the instructions to perform the steps of the method according to any one of the preceding claims.
Compared with the prior art, the invention has the beneficial effects that:
the image classification method and device supporting incremental learning provided by the invention can be used for classifying images on line, and learning can be carried out according to the incremental images, so that the classification accuracy is improved.
Drawings
Fig. 1 is a flowchart of a picture classification method supporting incremental learning according to an embodiment of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present invention, and are not intended to limit the scope of the present invention.
Embodiment one:
as shown in fig. 1, an embodiment of the present invention provides a method for classifying pictures, including the following steps:
step 1, selecting training picture samples in a training set, inputting the training picture samples into a convolutional neural network, and selecting typical picture samples through the convolutional neural network.
And step 2, updating the representative memory through the typical picture sample.
And step 3, carrying out data augmentation on the data set of the representative memory and the increment task to construct a test set.
And 4, inputting the test picture samples in the test set into a convolutional neural network, and extracting feature vectors through the convolutional neural.
And 5, inputting the feature vectors of all the test picture samples into a trained classifier, and outputting a picture classification result through the increment of the classifier.
Taking MNIST handwriting data set as an example, MNIST data set is a very classical data set in the machine learning field, and is composed of 60000 training samples and 10000 test samples, each sample is a 28 x 28 pixel gray scale handwriting digital picture. In the experiment, 10 kinds of samples from the number of 0 to 9 are selected, and 1000 pictures are selected as initial training set samples from each kind.
The selecting of the typical picture sample through the convolutional neural network comprises the following steps:
step 1.1, a convolutional neural network outputs posterior probability according to a training picture sample which can be correctly identified in input training picture samples;
and 1.2, selecting a typical picture sample from the training picture samples according to the sequence of the posterior probability from high to low.
Extracting feature vectors by convolving nerves includes:
step 4.1, carrying out convolution processing on an input test picture sample through a convolution layer to obtain image characteristics;
step 4.2, carrying out normalization processing on the image characteristics through batch normalization;
step 4.3, pooling the normalized image features through a MaxPooling layer;
and 4.4, activating the pooled image features through the ReLU activation function to obtain feature vectors.
Wherein the convolutional neural network comprises: 8 convolution layers, wherein each convolution layer uses a convolution kernel of 3×3 and 64 nodes, maxPooling layers are added after the 2 nd, 4 th, 6 th and 8 th convolution layers, the pool size of each MaxPooling layer is 2×2, and the step length is 2; and adding a Dropout layer on the last MaxPooling layer, wherein the Dropout probability is 0.5.
The classifier comprises a regular dual average algorithm RDA and a soft confidence weighting algorithm SCW, wherein the regular dual average algorithm RDA is used for improving the feature extraction capacity of the stream data, and the soft confidence weighting algorithm SCW is used for outputting classification results.
The classifier comprises a regular dual average algorithm RDA and a soft confidence weighting algorithm SCW, wherein the regular dual average algorithm RDA is used for improving the feature extraction capacity of the stream data, and the soft confidence weighting algorithm SCW is used for outputting classification results.
The expression of the optimization problem of the soft confidence weighting algorithm SCW is as follows:
where N (μ, Σ) represents the Gaussian distribution of the classifier obeying the mean vector μ and covariance matrix Σ, D KL Representing a new Gaussian distribution N (μ) t+1 ,Σ t+1 ) And the current Gaussian distribution N (mu) t ,Σ t ) Degree of divergence, iota between Φ As a loss function, C represents a trade-off between passive and aggressive parameters, (x) t ,y t ) Representing the input feature vector;
the sealing solution is as follows:
wherein alpha is t And beta t The coefficients, respectively, are expressed as follows:
wherein,m t =y t (μ t ·x t );Φ=Φ -1 (eta); phi represents a cumulative function of normal distribution, phi -1 (eta) is the inverse of phi, the inverseFor generating random variables obeying random distribution, eta e [0,1 ]]。
Preferably, the expression of the weight update strategy of the regular dual average algorithm RDA is as follows:
wherein,<G (t) ,W>representing gradient G (t) The integrated average value of the weight W, ψ (W) is a regular term, ψ (W) = lambda W 1 Lambda denotes the learning rate, h (W) is an auxiliary rigid convex function,β (t) is a non-negative and non-self-decreasing sequence, < >>t represents a time step.
Training of the classifier includes:
step 5.1, selecting training picture samples in a training set, inputting the training picture samples into a convolutional neural network, and extracting feature vectors through the convolutional neural network;
step 5.2, outputting a prediction label by the classifier according to the feature vector of the input training picture sample;
step 5.3, judging whether the predicted label is consistent with the actual label of the training picture sample,
if yes, the classifier is kept unchanged, and if not, parameters of the classifier are updated.
The invention herein compares to conventional incremental learning algorithms: the regular dual average algorithm and the soft confidence weighting algorithm have certain optimization on classification accuracy and model convergence speed, and are very practical picture classification methods for online classification of pictures.
Embodiment two:
the embodiment of the invention provides a picture classification device supporting incremental learning, which comprises a processor and a storage medium; the storage medium is used for storing instructions; the processor is operative according to instructions to perform steps of any of the methods according to the embodiments.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The foregoing is merely a preferred embodiment of the present invention, and it should be noted that modifications and variations could be made by those skilled in the art without departing from the technical principles of the present invention, and such modifications and variations should also be regarded as being within the scope of the invention.
Claims (6)
1. A picture classification method, comprising:
selecting training picture samples in a training set, inputting the training picture samples into a convolutional neural network, and selecting typical picture samples through the convolutional neural network;
updating the representative memory by a typical picture sample;
performing data augmentation on the data set of the representative memory and the increment task to construct a test set;
inputting the test picture samples in the test set into a convolutional neural network, and extracting feature vectors through the convolutional neural;
inputting the feature vectors of all the test picture samples into a trained classifier, and outputting a picture classification result through the increment of the classifier;
the classifier comprises a regular dual average algorithm RDA and a soft confidence weighting algorithm SCW, wherein the regular dual average algorithm RDA is used for improving the feature extraction capacity of stream data, and the soft confidence weighting algorithm SCW is used for outputting classification results;
the expression of the optimization problem of the soft confidence weighting algorithm SCW is as follows:
where N (μ, Σ) represents the Gaussian distribution of the classifier obeying the mean vector μ and covariance matrix Σ, D KL Representing a new Gaussian distribution N (μ) t+1 ,Σ t+1 ) And the current Gaussian distribution N (mu) t ,Σ t ) Degree of divergence, iota between Φ As a loss function, C represents a trade-off between passive and aggressive parameters, (x) t ,y t ) Representing the input feature vector;
the sealing solution is as follows:
wherein alpha is t And beta t The coefficients, respectively, are expressed as follows:
wherein,m t =y t (μ t ·x t ); Φ=Φ -1 (eta); phi represents a cumulative function of normal distribution, phi -1 (eta) is an inverse function of phi, which is used to generate random variables subject to random distribution, eta e [0,1];
The expression of the weight update strategy of the regular dual average algorithm RDA is as follows:
wherein,<G (t) ,W>representing gradient G (t) The integrated average value of the weight W, ψ (W) is a regular term, ψ (W) = lambda W 1 Lambda denotes the learning rate, h (W) is an auxiliary rigid convex function,β (t) is a non-negative and non-self-decreasing sequence, < >>t represents a time step.
2. The method of claim 1, wherein the selecting a representative picture sample via a convolutional neural network comprises:
the convolutional neural network outputs posterior probability according to training picture samples which can be correctly identified in the input training picture samples, and typical picture samples are selected from the training picture samples according to the sequence of the posterior probability from high to low.
3. A method of classifying pictures according to claim 2, wherein said convolutional neural network comprises: 8 convolution layers, wherein each convolution layer uses a convolution kernel of 3×3 and 64 nodes, maxPooling layers are added after the 2 nd, 4 th, 6 th and 8 th convolution layers, the pool size of each MaxPooling layer is 2×2, and the step length is 2; and adding a Dropout layer on the last MaxPooling layer, wherein the Dropout probability is 0.5.
4. A method of classifying pictures according to claim 3, wherein said extracting feature vectors by convolutional nerves comprises:
carrying out convolution processing on an input test picture sample through a convolution layer to obtain image characteristics;
normalizing the image features through batch normalization;
carrying out pooling treatment on the normalized image features through a MaxPooling layer;
and activating the pooled image features through a ReLU activation function to obtain feature vectors.
5. A method of classifying pictures according to claim 1, wherein the training of the classifier comprises:
selecting training picture samples in a training set, inputting the training picture samples into a convolutional neural network, and extracting feature vectors through the convolutional neural network;
the classifier outputs a prediction label according to the feature vector of the input training picture sample;
judging whether the predicted label is consistent with the actual label of the training picture sample,
if yes, the classifier is kept unchanged, and if not, parameters of the classifier are updated.
6. The picture classifying device is characterized by comprising a processor and a storage medium;
the storage medium is used for storing instructions;
the processor being operative according to the instructions to perform the steps of the method according to any one of claims 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110716563.0A CN113591913B (en) | 2021-06-28 | 2021-06-28 | Picture classification method and device supporting incremental learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110716563.0A CN113591913B (en) | 2021-06-28 | 2021-06-28 | Picture classification method and device supporting incremental learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113591913A CN113591913A (en) | 2021-11-02 |
CN113591913B true CN113591913B (en) | 2024-03-29 |
Family
ID=78244701
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110716563.0A Active CN113591913B (en) | 2021-06-28 | 2021-06-28 | Picture classification method and device supporting incremental learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113591913B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
CN107967497A (en) * | 2017-08-28 | 2018-04-27 | 苏州明逸智库信息科技有限公司 | Manuscripted Characters Identification Method based on convolutional neural networks and extreme learning machine |
CN111222545A (en) * | 2019-12-24 | 2020-06-02 | 西安电子科技大学 | Image classification method based on linear programming incremental learning |
CN111488917A (en) * | 2020-03-19 | 2020-08-04 | 天津大学 | Garbage image fine-grained classification method based on incremental learning |
CN112115973A (en) * | 2020-08-18 | 2020-12-22 | 吉林建筑大学 | Convolutional neural network based image identification method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108932480B (en) * | 2018-06-08 | 2022-03-15 | 电子科技大学 | Distributed optical fiber sensing signal feature learning and classifying method based on 1D-CNN |
-
2021
- 2021-06-28 CN CN202110716563.0A patent/CN113591913B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
CN107967497A (en) * | 2017-08-28 | 2018-04-27 | 苏州明逸智库信息科技有限公司 | Manuscripted Characters Identification Method based on convolutional neural networks and extreme learning machine |
CN111222545A (en) * | 2019-12-24 | 2020-06-02 | 西安电子科技大学 | Image classification method based on linear programming incremental learning |
CN111488917A (en) * | 2020-03-19 | 2020-08-04 | 天津大学 | Garbage image fine-grained classification method based on incremental learning |
CN112115973A (en) * | 2020-08-18 | 2020-12-22 | 吉林建筑大学 | Convolutional neural network based image identification method |
Non-Patent Citations (1)
Title |
---|
一种基于局部线性嵌入的SVM增量学习方法;姚明海 等;浙江工业大学学报;20190531;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN113591913A (en) | 2021-11-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11521095B2 (en) | Methods and systems for CNN network adaption and object online tracking | |
US11941523B2 (en) | Stochastic gradient boosting for deep neural networks | |
CN107506799B (en) | Deep neural network-based mining and expanding method and device for categories of development | |
EP3767536A1 (en) | Latent code for unsupervised domain adaptation | |
CN113077388B (en) | Data-augmented deep semi-supervised over-limit learning image classification method and system | |
CN112446423A (en) | Fast hybrid high-order attention domain confrontation network method based on transfer learning | |
CN113159072B (en) | Online ultralimit learning machine target identification method and system based on consistency regularization | |
KR20190004429A (en) | Method and apparatus for determining training of unknown data related to neural networks | |
EP4020338A1 (en) | Information processing apparatus and information processing method | |
CN111652264B (en) | Negative migration sample screening method based on maximum mean value difference | |
Wang et al. | Jumpout: improved dropout for deep neural networks with ReLUs | |
Seyedhosseini et al. | Fast adaboost training using weighted novelty selection | |
CN116912568A (en) | Noise-containing label image recognition method based on self-adaptive class equalization | |
Zhuang et al. | A handwritten Chinese character recognition based on convolutional neural network and median filtering | |
Tao et al. | Efficient incremental training for deep convolutional neural networks | |
CN111062465A (en) | Image recognition model and method with neural network structure self-adjusting function | |
CN113591913B (en) | Picture classification method and device supporting incremental learning | |
CN112287662A (en) | Natural language processing method, device and equipment based on multiple machine learning models | |
CN111753995A (en) | Local interpretable method based on gradient lifting tree | |
CN108573275B (en) | Construction method of online classification micro-service | |
Kang et al. | Comparison of weight initialization techniques for deep neural networks | |
JP2022075620A (en) | Methods and systems for training convolutional neural networks | |
US20210365719A1 (en) | System and method for few-shot learning | |
CN114202006A (en) | Neural network selection | |
Swarnkar et al. | A paradigm shift for computational excellence from traditional machine learning to modern deep learning-based image steganalysis |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |