CN108830380A - A kind of training pattern generation method and system based on cloud service - Google Patents
A kind of training pattern generation method and system based on cloud service Download PDFInfo
- Publication number
- CN108830380A CN108830380A CN201810322042.5A CN201810322042A CN108830380A CN 108830380 A CN108830380 A CN 108830380A CN 201810322042 A CN201810322042 A CN 201810322042A CN 108830380 A CN108830380 A CN 108830380A
- Authority
- CN
- China
- Prior art keywords
- training
- cloud server
- client
- generation method
- object module
- 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.)
- Granted
Links
Classifications
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Electrically Operated Instructional Devices (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of training pattern generation method and system based on cloud service, retraining is carried out on the basis of case mold using the training object that Cloud Server is submitted for user, get the network model sensitive to specific objective, optimization case mold is realized to the sensitivity of training object, ignore other targets, further model is compressed, the model after training with making is readily able to be deployed in embedded environment, and model is enabled to make inferences and predict locally.
Description
Technical field
The present invention relates to the artificial intelligence fields of neural network model and deep learning, more particularly to one kind to be based on cloud service
Training pattern generation method and system.
Background technique
In recent years, with artificial intelligence (AI：Artificial Intelligence) fast development, deep learning
(DL:Deep Learning) network due to by combination low-level image feature formed high-level characteristic, influenced by environmental change it is smaller,
Breakthrough achievement is achieved in computer vision field.
Currently, existing deep learning model needs are constantly trained according to the update of training dataset, obtain more
Accurate prediction model, by being updated the prediction model to the deep learning model, so that the prediction model is used for
It is online to execute data predicted operation.However, AI application is normally based on embedded end in artificial intelligence practical application, pass through
Off-line model is locally making inferences, is predicting.And different AI applications are for different execution objects there are different models,
It needs specifically to be trained.
The prior art mainly solves the on-line study ability of network model, specifically includes and submits certain training data
Collection, then obtains more accurate prediction model, does not accomplish classification, detection or identification only to certain this target, can not also do
To compressing and be deployed in embedded environment.
Summary of the invention
For the above-mentioned problems in the prior art, now provide a kind of training pattern generation method based on cloud service and
System.
Specific technical solution is as follows：
A kind of training pattern generation method based on cloud service, the Cloud Server remotely connect local by network interface
Client, be provided with multiclass case mold in the Cloud Server；
Include the following steps：
Step S1：User uploads preset training data to the Cloud Server by the client, and selects
This trains the required case mold；
Step S2：The Cloud Server uses the training data to be trained to obtain a target mould case mold
Type；
Step S3：The Cloud Server carries out weighed value adjusting to the object module；
Step S4：The Cloud Server is tested for the property the object module after weighed value adjusting, and according to
Test result judges whether the object module reaches preset performance indicator；
If so, entering step S6；If it is not, then entering step S5；
Step S5：The Cloud Server carries out weighed value adjusting to the object module again, is then returned to the step S4；
Step S6：The Cloud Server saves the training pattern that the object module is needed as user, and to
Training pattern described in the client feedback.
Preferably, the training data includes training objective；
It is provided with a database in the Cloud Server, multiple local data sets, Mei Gesuo are prestored in the database
It states local data sets and corresponds respectively at least one described training objective, each local data sets correspond respectively at least one
Case mold described in class；
The step S1 is specifically included：
Step S11a：The user uploads the training objective to the Cloud Server by the client；
Step S12a：The Cloud Server retrieves to obtain and match with the training objective database
The local data sets；
If not retrieving the matched local data sets, S13a is entered step；
If retrieving the matched local data sets, step S14a；
Step S13a：To the message of the client feedback failure to train, with backed off after random；
Step S14a：The message being properly received to the client feedback, and receive that the client sends starts to instruct
Experienced training instruction, the local data sets for then obtaining retrieval turn to the step S2 as the training data.
Preferably, the training data includes training dataset；
The step S1 is specifically included：
Step S11b, the user upload the training dataset to the Cloud Server by the client；
Step S12b successfully connects after the Cloud Server is properly received the training dataset to the client feedback
The message of receipts；
Step S13b, the user start trained training instruction to Cloud Server transmission by the client,
Simultaneous selection this train the required case mold；
Then in the step S2, the Cloud Server uses the training dataset, to the mother of user selection
Model is trained.
Preferably, weight tune is carried out using following steps to the object module in the step S3 and step S5
It is whole：
Step A1：Count the change frequency and variable quantity of each weight in the object module generating process；
Step A2：The change frequency and the corresponding variable quantity are compared with the preset value to obtain variation
The lesser weight of amplitude；
Step A3：Preset parameter is set by the lesser weight of the middle amplitude of variation of the object module and reduces institute
State the output characteristic quantity of object module.
Preferably, the preset value includes frequency threshold value and change threshold；
The step A2 includes the following steps：
Step A21：Change frequency described in each weight is compared with the frequency threshold value respectively, and will be described
Change frequency is less than the weight of the frequency threshold value as intermediate weight；
Step A22：The variable quantity of each intermediate weight is compared with the change threshold respectively, in the variation
When amount is respectively less than the frequency threshold value, the variable quantity is obtained corresponding to the initial weight in the case mold, and will be described first
Beginning weight is as the lesser weight of amplitude of variation.
Preferably, in the step S5, after being adjusted to the preset value, then weighed value adjusting is carried out.
Preferably, further comprising the steps of in the step S5：
The adjustment number being adjusted to the preset value is counted, and after the adjustment number is greater than preset number,
To the message of the client feedback failure to train, with backed off after random.
Preferably, the case mold includes multiple convolutional layers and multiple full articulamentums；
In the step S2, the learning functionality of the part convolutional layer is closed, start the convolutional layer of remainder and is owned
The full articulamentum learning functionality.
Preferably, a kind of training pattern based on cloud service generates system, raw using any of the above-described training pattern
At method.
Above-mentioned technical proposal has the following advantages that or beneficial effect：
Retraining is carried out on the basis of case mold using the training data that Cloud Server is submitted for user, is got pair
The network model of specific objective sensitivity realizes optimization case mold to the sensitivity of training data, ignores other targets, into one
Step compresses model, and the model after training with making is readily able to be deployed in embedded environment, enables model at this
Ground makes inferences and predicts.
Detailed description of the invention
With reference to appended attached drawing, more fully to describe the embodiment of the present invention.However, appended attached drawing be merely to illustrate and
It illustrates, and is not meant to limit the scope of the invention.
Fig. 1 is a kind of flow chart of the embodiment of the training pattern generation method based on cloud service of the present invention；
Fig. 2 is the flow chart that training objective is uploaded in the embodiment of the present invention；
Fig. 3 is the flow chart that training dataset is uploaded in the embodiment of the present invention；
Fig. 4 is the flow chart that weighed value adjusting is carried out in the embodiment of the present invention；
Fig. 5 is the flow chart that the lesser weight of amplitude of variation is obtained in the embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art without creative labor it is obtained it is all its
His embodiment, shall fall within the protection scope of the present invention.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the present invention can phase
Mutually combination.
The present invention will be further explained below with reference to the attached drawings and specific examples, but not as the limitation of the invention.
In a kind of preferred embodiment of the present invention, according to Fig. 1, a kind of training pattern generation side based on cloud service
Method, Cloud Server remotely connect local client by network interface, multiclass case mold are provided in Cloud Server；
Include the following steps：
Step S1：User uploads preset training data to Cloud Server by client, and selects this training
Required case mold；
Step S2：Cloud Server uses training data to be trained case mold to obtain an object module；
Step S3：Cloud Server carries out weighed value adjusting to object module；
Step S4：Cloud Server is tested for the property the object module after weighed value adjusting, and according to test result
Judge whether object module reaches preset performance indicator；
If so, entering step S6；If it is not, then entering step S5；
Step S5：Cloud Server carries out weighed value adjusting to object module again, is then returned to step S4；
Step S6：Cloud Server saves the training pattern that object module is needed as user, and anti-to client
Present training pattern.
Specifically, in the present embodiment, the training data of user's submission is directed on the basis of case mold using Cloud Server
Retraining is carried out, implementation model is sensitive to specific objective,
User uploads training data according to demand, and suitable mother is chosen according to the particular content of training data in Cloud Server
The training that model is directed to.Case mold includes but is not limited to generic object detection model and object classification model.Case mold
Optimization is realized based on different data set analysis study, and training data is that data set then directly carries out depth in the present embodiment
Study when training data is specific target, carries out depth to case mold by transferring corresponding data set in Cloud Server
Degree study.
All kinds of case molds (Base Model) of deep learning are provided in Cloud Server, and establish the network for carrying out cloud service
Interface, including data set submit (Dataset Upload), training (ReTraining), training monitoring (Monitor
Training), network model downloading (Model_Download).Cloud Server and client are realized by above-mentioned network interface
Interaction.Network interface uses the form of api interface.
For user in submitting interface to upload training data by data set in client, training data includes the instruction of user oneself
The target or requirement that white silk data set or user specify.The set about certain target objects that training dataset refers to.For
Oneself training dataset then by indicating specific target or requirement does not carry out model training to user.
Cloud Server receives submits interface to client feedback reception state after training data by data set.User in
After acquisition of information in client by reception state uploads successfully to training data, Cloud Server is controlled by teaching interface and is opened
Dynamic training.
In above-described embodiment, first to case mold be based on training data carry out deep learning training, obtain object module with
And in training process each weight situation of change；Further, according to the preset value as threshold value (Threshold) to weight
Situation of change is compared analysis, obtains and changes small weight；Thereafter, the small weight of above-mentioned variation is directly set to default
It is worth (NA), while reduces output characteristic quantity, obtains object module adjusted, and analysis is carried out to the performance of object module and is sentenced
It is disconnected, it is unsatisfactory for desired object module for performance, preset value is adjusted, until object module meets performance requirement.
Cloud Server is saved using trained object module as training result, notifies client by network interface
End, client pass through the model that the downloading training of network model download interface is completed again.
In a kind of preferred embodiment of the present invention, according to Fig.2, training data includes training objective；
It is provided with a database in Cloud Server, multiple local data sets, each local data sets are prestored in database
At least one training objective is corresponded respectively to, each local data sets correspond respectively at least a kind of case mold；
Step S1 is specifically included：
Step S11a：User uploads training objective to Cloud Server by client；
Step S12a：Cloud Server retrieves to obtain the local data sets to match with training objective database；
If not retrieving matched local data sets, S13a is entered step；
If retrieving matched local data sets, step S14a；
Step S13a：To the message of client feedback failure to train, with backed off after random；
Step S14a：The message being properly received to client feedback, and receive the training for starting training of client transmission
Instruction, the local data sets for then obtaining retrieval turn to step S2 as training data.
According to Fig.3, training data includes training dataset；
Step S1 is specifically included：
Step S11b：User uploads training dataset to Cloud Server by client；
Step S12b：After Cloud Server is properly received training dataset, to the message of client feedback successful reception；
Step S13b：User starts trained training instruction to Cloud Server transmission by client, simultaneous selection this
The required case mold of training；
Then in step S2, Cloud Server uses training dataset, is trained to the case mold of user's selection.
Specifically, in the present embodiment, the training data that Cloud Server receives is training objective, searches for local data set,
With the presence or absence of the data set of user's requirement is met, retraining is carried out based on case mold if so, then beginning preparing；If no, ringing
Such target or object retraining service should can not be provided to user.
User is submitting the training data sent when deep learning request, can be the data set of user's offer, can be
Perhaps specific sensitive objects above scheme is carried out to specific target or specific sensitive objects to a kind of specific target
Matching, obtains suitable training material by matching to the data set inside Cloud Server, trains to reach and upload with user
The identical effect of data set.
In a kind of preferred embodiment of the present invention, object classification model is carried out for training, Yong Huti in Cloud Server
The training objective of friendship is fruit identification, is stored with the other local data sets of fruits for Cloud Server, then feeds back and upload successfully.
It is not stored with the other local data sets of fruits for Cloud Server, then feeds back upload failure, needs user on the client
The other training dataset of fruits is passed, just can be carried out the training to case mold.
In a kind of preferred embodiment of the present invention, case mold includes multiple convolutional layers and multiple full articulamentums；
In step S2, in executing deep learning training process, the learning functionality of close portion bundling lamination starts remainder
The learning functionality of the convolutional layer and all full articulamentums that divide.
In a kind of preferred embodiment of the present invention, in executing deep learning training process, case mold front half section volume is closed
The learning functionality of lamination, the learning functionality of two layers of convolutional layer and the learning functionality of all full articulamentums after reservation.
In a kind of preferred embodiment of the present invention, according to Fig.4, in step S3 and step S5 to object module use with
Lower step carries out weighed value adjusting：
Step A1：Count the change frequency and variable quantity of each weight in object module generating process；
Step A2：Change frequency and corresponding variable quantity are compared with preset value to obtain the lesser power of amplitude of variation
Value；
Step A3：Preset parameter is set by the lesser weight of middle amplitude of variation of object module and reduces object module
Export characteristic quantity.
Specifically, in the present embodiment, by the way that the weight and output characteristic quantity of preset value controlled training model is arranged, pressure is realized
Contracting model make training after model be readily able to be deployed in embedded environment, enable model locally make inferences and
Prediction.
In a kind of preferred embodiment of the present invention, according to Fig.5, preset value includes frequency threshold value and change threshold；
Step A2 includes the following steps：
Step A21：Each weight change frequency is compared with frequency threshold value respectively, and change frequency is less than number
The weight of threshold value is as intermediate weight；
Step A22：The variable quantity of each intermediate weight is compared with change threshold respectively, is respectively less than number in variable quantity
When threshold value, the initial weight that variable quantity corresponds in case mold is obtained, and using initial weight as the lesser weight of amplitude of variation.
In a kind of preferred embodiment of the present invention, in step S3, after i-th iteration training, statistics opens study function
Can layer weight situation of change, obtain w1 weight, w2 weight ..., the change frequency of wi weight ... is n1, n2 ... ni ... and
Variation delta 1, Δ 2 ... Δ i ....By change frequency and variable quantity by sorting from small to large, change frequency and variable quantity are obtained
Ultimate sequence.Using the scheme of above-described embodiment, obtains the lesser weight of amplitude of variation in ultimate sequence and be directly set to NA, no
Calculating is participated in, while reducing output characteristic quantity.
In a kind of preferred embodiment of the present invention, in step S5, after being adjusted to preset value, then weight tune is carried out
It is whole.
Using by adjusting preset value, the object of weighed value adjusting can be changed in the above scheme, to realize to target
The adjustment of model.
It is further comprising the steps of in step S5 in a kind of preferred embodiment of the present invention：
The adjustment number being adjusted to preset value is counted, and after adjustment number is greater than preset number, to client
Feedback training failure news, with backed off after random.
The adjustment of preset value is used in step S5, frequency threshold value and change threshold are increased or reduced respectively.
Specifically, in the present embodiment, it is so unable to get expected training result for repeatedly adjusting preset value times, by upper
The scheme stated stops the optimization to model, issues failed message to client.
In a kind of preferred embodiment of the present invention, a kind of training pattern generation system based on cloud service, using above-mentioned
Training pattern generation method described in one.
The foregoing is merely preferred embodiments of the present invention, are not intended to limit embodiments of the present invention and protection model
It encloses, to those skilled in the art, should can appreciate that all with made by description of the invention and diagramatic content
Equivalent replacement and obviously change obtained scheme, should all be included within the scope of the present invention.
Claims (9)
1. a kind of training pattern generation method based on cloud service, which is characterized in that the Cloud Server is remote by network interface
Journey connects local client, is provided with multiclass case mold in the Cloud Server；
Include the following steps：
Step S1：User uploads preset training data to the Cloud Server by the client, and selects this
The required case mold of training；
Step S2：The Cloud Server uses the training data to be trained the case mold to obtain an object module；
Step S3：The Cloud Server carries out weighed value adjusting to the object module；
Step S4：The Cloud Server is tested for the property the object module after weighed value adjusting, and according to test
As a result judge whether the object module reaches preset performance indicator；
If so, entering step S6；If it is not, then entering step S5；
Step S5：The Cloud Server carries out weighed value adjusting to the object module again, is then returned to the step S4；
Step S6：The Cloud Server saves the training pattern that the object module is needed as user, and to described
Training pattern described in client feedback.
2. training pattern generation method according to claim 1, which is characterized in that the training data includes training mesh
Mark；
It is provided with a database in the Cloud Server, prestores multiple local data sets in the database, each described
Ground data set corresponds respectively at least one described training objective, and each local data sets correspond respectively at least a kind of institute
State case mold；
The step S1 is specifically included：
Step S11a：The user uploads the training objective to the Cloud Server by the client；
Step S12a：The Cloud Server is retrieved the database with described in obtaining and matching with the training objective
Local data sets；
If not retrieving the matched local data sets, S13a is entered step；
If retrieving the matched local data sets, step S14a；
Step S13a：To the message of the client feedback failure to train, with backed off after random；
Step S14a：The message being properly received to the client feedback, and receive that the client sends starts training
Training instruction, the local data sets for then obtaining retrieval turn to the step S2 as the training data.
3. training pattern generation method according to claim 1, which is characterized in that the training data includes training data
Collection；
The step S1 is specifically included：
Step S11b, the user upload the training dataset to the Cloud Server by the client；
Step S12b after the Cloud Server is properly received the training dataset, is properly received to the client feedback
Message；
Step S13b, the user start trained training instruction to Cloud Server transmission by the client, simultaneously
This is selected to train the required case mold；
Then in the step S2, the Cloud Server uses the training dataset, to the case mold of user selection
It is trained.
4. training pattern generation method according to claim 1, which is characterized in that in the step S3 and the step S5
Weighed value adjusting is carried out using following steps to the object module：
Step A1：Count the change frequency and variable quantity of each weight in the object module generating process；
Step A2：The change frequency and the corresponding variable quantity are compared with the preset value to obtain amplitude of variation
The lesser weight；
Step A3：Preset parameter is set by the lesser weight of the middle amplitude of variation of the object module and reduces the mesh
Mark the output characteristic quantity of model.
5. training pattern generation method according to claim 4, which is characterized in that the preset value include frequency threshold value and
Change threshold；
The step A2 includes the following steps：
Step A21：Change frequency described in each weight is compared with the frequency threshold value respectively, and by the variation
Number is less than the weight of the frequency threshold value as intermediate weight；
Step A22：The variable quantity of each intermediate weight is compared with the change threshold respectively, equal in the variable quantity
When less than the frequency threshold value, the variable quantity is obtained corresponding to the initial weight in the case mold, and by the initial power
Value is used as the lesser weight of amplitude of variation.
6. training pattern generation method according to claim 4, which is characterized in that in the step S5, using to described
After preset value is adjusted, then carry out weighed value adjusting.
7. training pattern generation method according to claim 6, which is characterized in that further include following step in the step S5
Suddenly：
The adjustment number being adjusted to the preset value is counted, and after the adjustment number is greater than preset number, to institute
The message for stating client feedback failure to train, with backed off after random.
8. training pattern generation method according to claim 1, which is characterized in that the case mold includes multiple convolutional layers
With multiple full articulamentums；
In the step S2, close the part convolutional layer learning functionality, start remainder convolutional layer and all institutes
State the learning functionality of full articulamentum.
9. a kind of training pattern based on cloud service generates system, which is characterized in that using as described in any in claim 1-8
Training pattern generation method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810322042.5A CN108830380B (en) | 2018-04-11 | 2018-04-11 | Training model generation method and system based on cloud service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810322042.5A CN108830380B (en) | 2018-04-11 | 2018-04-11 | Training model generation method and system based on cloud service |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108830380A true CN108830380A (en) | 2018-11-16 |
| CN108830380B CN108830380B (en) | 2021-10-01 |
Family
ID=64155481
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810322042.5A Active CN108830380B (en) | 2018-04-11 | 2018-04-11 | Training model generation method and system based on cloud service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108830380B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109614777A (en) * | 2018-11-23 | 2019-04-12 | 第四范式（北京）技术有限公司 | The user ID authentication method and device of smart machine and smart machine |
| CN109657799A (en) * | 2018-11-20 | 2019-04-19 | 北京首钢自动化信息技术有限公司 | A kind of model tuning method and apparatus based on scene adaptation |
| CN110443149A (en) * | 2019-07-10 | 2019-11-12 | 安徽万维美思信息科技有限公司 | Target object searching method, system and storage medium |
| WO2020177377A1 (en) * | 2019-03-07 | 2020-09-10 | 平安科技（深圳）有限公司 | Machine learning-based data prediction processing method and apparatus, and computer device |
| CN111709522A (en) * | 2020-05-21 | 2020-09-25 | 哈尔滨工业大学 | Deep learning target detection system based on server-embedded cooperation |
| CN111796840A (en) * | 2019-04-09 | 2020-10-20 | Oppo广东移动通信有限公司 | Algorithm model updating method and device, storage medium and electronic equipment |
| WO2021139465A1 (en) * | 2020-01-09 | 2021-07-15 | 深圳前海微众银行股份有限公司 | Backward model selection method and device, and readable storage medium |
| CN113222121A (en) * | 2021-05-31 | 2021-08-06 | 杭州海康威视数字技术股份有限公司 | Data processing method, device and equipment |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102983819A (en) * | 2012-11-08 | 2013-03-20 | 南京航空航天大学 | Imitating method of power amplifier and imitating device of power amplifier |
| CN105095322A (en) * | 2014-05-23 | 2015-11-25 | 富士通株式会社 | Personnel name unit dictionary expansion method, personnel name language recognition method, personnel name unit dictionary expansion device and personnel name language recognition device |
| CN105095962A (en) * | 2015-07-27 | 2015-11-25 | 中国汽车工程研究院股份有限公司 | Method for predicting dynamic mechanical property of material based on BP artificial neural network |
| CN106041912A (en) * | 2016-06-16 | 2016-10-26 | 深圳先进技术研究院 | Master-slave mode snake-like robot system and position control method thereof |
| CN106203417A (en) * | 2016-07-12 | 2016-12-07 | 北京科技大学 | A kind of adhesion character alienable RMB crown word number identification method |
| CN106295874A (en) * | 2016-08-08 | 2017-01-04 | 上海交通大学 | Traffic flow parameter Forecasting Methodology based on deep belief network |
| CN107145395A (en) * | 2017-07-04 | 2017-09-08 | 北京百度网讯科技有限公司 | Method and apparatus for handling task |
| CN107195186A (en) * | 2017-06-07 | 2017-09-22 | 千寻位置网络有限公司 | The method and system of Automatic Optimal crossing vehicle pass-through speed |
| CN107291232A (en) * | 2017-06-20 | 2017-10-24 | 深圳市泽科科技有限公司 | A kind of somatic sensation television game exchange method and system based on deep learning and big data |
| CN107316001A (en) * | 2017-05-31 | 2017-11-03 | 天津大学 | Small and intensive method for traffic sign detection in a kind of automatic Pilot scene |
| CN107506941A (en) * | 2017-09-09 | 2017-12-22 | 杭州数立方征信有限公司 | A kind of enterprise in charge of construction's credit assessment method and system based on big data technology |
-
2018
- 2018-04-11 CN CN201810322042.5A patent/CN108830380B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102983819A (en) * | 2012-11-08 | 2013-03-20 | 南京航空航天大学 | Imitating method of power amplifier and imitating device of power amplifier |
| CN105095322A (en) * | 2014-05-23 | 2015-11-25 | 富士通株式会社 | Personnel name unit dictionary expansion method, personnel name language recognition method, personnel name unit dictionary expansion device and personnel name language recognition device |
| CN105095962A (en) * | 2015-07-27 | 2015-11-25 | 中国汽车工程研究院股份有限公司 | Method for predicting dynamic mechanical property of material based on BP artificial neural network |
| CN106041912A (en) * | 2016-06-16 | 2016-10-26 | 深圳先进技术研究院 | Master-slave mode snake-like robot system and position control method thereof |
| CN106203417A (en) * | 2016-07-12 | 2016-12-07 | 北京科技大学 | A kind of adhesion character alienable RMB crown word number identification method |
| CN106295874A (en) * | 2016-08-08 | 2017-01-04 | 上海交通大学 | Traffic flow parameter Forecasting Methodology based on deep belief network |
| CN107316001A (en) * | 2017-05-31 | 2017-11-03 | 天津大学 | Small and intensive method for traffic sign detection in a kind of automatic Pilot scene |
| CN107195186A (en) * | 2017-06-07 | 2017-09-22 | 千寻位置网络有限公司 | The method and system of Automatic Optimal crossing vehicle pass-through speed |
| CN107291232A (en) * | 2017-06-20 | 2017-10-24 | 深圳市泽科科技有限公司 | A kind of somatic sensation television game exchange method and system based on deep learning and big data |
| CN107145395A (en) * | 2017-07-04 | 2017-09-08 | 北京百度网讯科技有限公司 | Method and apparatus for handling task |
| CN107506941A (en) * | 2017-09-09 | 2017-12-22 | 杭州数立方征信有限公司 | A kind of enterprise in charge of construction's credit assessment method and system based on big data technology |
Non-Patent Citations (3)
| Title |
|---|
| ZHUANG LIU ET AL.: "Learning Efficient Convolutional Networks through Network Slimming", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| 周维华: "RBF神经网络隐层结构与参数优化研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 张江: "吐血推荐：超级好用的深度学习云平台Floyd", 《HTTPS://ZHUANLAN.ZHIHU.COM/P/27573615》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109657799A (en) * | 2018-11-20 | 2019-04-19 | 北京首钢自动化信息技术有限公司 | A kind of model tuning method and apparatus based on scene adaptation |
| CN109614777A (en) * | 2018-11-23 | 2019-04-12 | 第四范式（北京）技术有限公司 | The user ID authentication method and device of smart machine and smart machine |
| CN109614777B (en) * | 2018-11-23 | 2020-09-11 | 第四范式（北京）技术有限公司 | Intelligent device and user identity authentication method and device of intelligent device |
| WO2020177377A1 (en) * | 2019-03-07 | 2020-09-10 | 平安科技（深圳）有限公司 | Machine learning-based data prediction processing method and apparatus, and computer device |
| CN111796840A (en) * | 2019-04-09 | 2020-10-20 | Oppo广东移动通信有限公司 | Algorithm model updating method and device, storage medium and electronic equipment |
| CN110443149A (en) * | 2019-07-10 | 2019-11-12 | 安徽万维美思信息科技有限公司 | Target object searching method, system and storage medium |
| WO2021139465A1 (en) * | 2020-01-09 | 2021-07-15 | 深圳前海微众银行股份有限公司 | Backward model selection method and device, and readable storage medium |
| CN111709522A (en) * | 2020-05-21 | 2020-09-25 | 哈尔滨工业大学 | Deep learning target detection system based on server-embedded cooperation |
| CN111709522B (en) * | 2020-05-21 | 2022-08-02 | 哈尔滨工业大学 | Deep learning target detection system based on server-embedded cooperation |
| CN113222121A (en) * | 2021-05-31 | 2021-08-06 | 杭州海康威视数字技术股份有限公司 | Data processing method, device and equipment |
| CN113222121B (en) * | 2021-05-31 | 2023-08-29 | 杭州海康威视数字技术股份有限公司 | Data processing method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108830380B (en) | 2021-10-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108830380A (en) | A kind of training pattern generation method and system based on cloud service | |
| CN109902832A (en) | Training method, predicting abnormality method and the relevant apparatus of machine learning model | |
| CN108361927A (en) | A kind of air-conditioner control method, device and air conditioner based on machine learning | |
| CN105487500A (en) | Streaming data for analytics in process control systems | |
| CN109508376A (en) | It can online the error correction intension recognizing method and device that update | |
| CN110889509B (en) | Gradient momentum acceleration-based joint learning method and device | |
| WO2017066185A1 (en) | Performance adjustment for wireless power transfer devices | |
| CN106453608B (en) | A kind of background request adaptive scheduling algorithm of the mobile application based on cloud | |
| CN110326008A (en) | Machine learning is integrated into control system | |
| CN108520301A (en) | A kind of circuit intermittent fault diagnostic method based on depth confidence network | |
| CN107800572A (en) | A kind of method and apparatus based on neutral net updating apparatus | |
| CN104573185B (en) | The processing method and processing device of High-sped Trains index | |
| CN110324170A (en) | Data analysis equipment, multi-model are total to decision system and method | |
| CN105094305A (en) | Method for identifying user behavior, user equipment and behavior identification server | |
| CN110069715A (en) | A kind of method of information recommendation model training, the method and device of information recommendation | |
| CN109787699B (en) | Wireless sensor network routing link state prediction method based on mixed depth model | |
| CN115130065B (en) | Method, device and equipment for processing characteristic information of supply terminal and computer readable medium | |
| CN106446354B (en) | Hybrid electric vehicle power train system torsional oscillation prediction and removing method | |
| WO2022002068A1 (en) | Data processing method, system and device and storage medium | |
| CN109214515A (en) | A kind of deep neural network inference method and calculate equipment | |
| CN114238269A (en) | Database parameter adjusting method and device, electronic equipment and storage medium | |
| CN110175655A (en) | Data identification method and device, storage medium and electronic equipment | |
| CN112199154B (en) | Reinforced learning training system and method based on distributed collaborative sampling center type optimization | |
| CN110210881A (en) | A kind of user interest prediction technique, apparatus and system | |
| CN112799949A (en) | Model optimization method, device, system and equipment |
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 |