CN107977712A - Network type machine learning system - Google Patents
Network type machine learning system Download PDFInfo
- Publication number
- CN107977712A CN107977712A CN201711387686.4A CN201711387686A CN107977712A CN 107977712 A CN107977712 A CN 107977712A CN 201711387686 A CN201711387686 A CN 201711387686A CN 107977712 A CN107977712 A CN 107977712A
- Authority
- CN
- China
- Prior art keywords
- model
- equipment
- machine learning
- data
- parameter
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Debugging And Monitoring (AREA)
Abstract
The invention discloses network type machine learning system, the system comprises：Each independent machine learning model is run in the equipment of interconnection in system, when the model in an equipment in system is trained or tests, model, model parameter and the data that other equipment provides in the equipment frame of reference, the performance of local model is improved, or in the case of no local model, the model that is directly provided using other models, parameter, data, establish local model；In system each equipment can be invited or active upload local used in model, parameter, data；The local model of devices in system can actively provide test data, seek that the model and model parameter of test can be completed from system miscellaneous equipment, improve the speed of study, same model is subjected to the inspection of various application scenarios and data so as to improve performance.
Description
Technical field
The present invention relates to machine learning field, and in particular, to a kind of network type machine learning system.
Background technology
Using deep learning as the just fashionable whole world of the artificial intelligence technology of new generation of representative, become new air port, and it is extensive
Applied to fields such as image, natural language, financial instruments.But whether based on depth model or traditional shallow Model, all
According to data acquisition, local training, the flow tested, model is finally obtained, then solved practical problems with this model.Generally all
One training mission, even if being distributed formula training, is also simply assigned on more machines by the only training on an independent machine,
Lack automatic foreign exchanges mechanism in whole process.
Therefore, existing machine learning system is more closed, and pace of learning is slow, is not joined between each training machine
It is, there are repetition training between a large amount of machines, to be unfavorable for the raising of each machine learning model ability, and the training of individual machine
It is often relatively simple with data source, it have impact on its generalization ability.
The content of the invention
The present invention provides a kind of network type machine learning system, and solving existing machine learning, there are pace of learning
Slowly, it is not in contact between each machine, the problem of performance is relatively low, generalization ability is poor, improves the speed of study, can also make same mould
Type is subjected to the inspection of various application scenarios and data so as to improve performance.
For achieving the above object, this application provides network type machine learning system, the system comprises：
The respective independent machine learning model of equipment operation of interconnection in system.When in an equipment in system
Model when being trained or testing, equipment can directly acquire the machine learning model, model parameter, number of other equipment offer
According to improving to the performance of local model, also can directly be provided in the case of no local model using other models
Model, parameter, data, establish local model.Specific mode, one kind be equipment be invited or active upload local used in mould
Type, parameter or data, take for other equipment.Another situation is then that local model actively provides data, and seeking can be preferably complete
Into the model and model parameter of target.
Wherein, it will be distributed over different location model and carry out on-line joining process, each equipment passes through agreement, the directly local training of transmission
The model parameter gone out gives other models, is used for reference for other models, can be greatly enhanced the speed of study, can also make same
Model is subjected to the inspection of various application scenarios and data so as to improve performance.
Further, the system also includes the first management platform, each equipment to be uploaded to first by respective model, parameter
Management platform, and provide training and test data to the first management platform；Distinct device according to itself application scenarios and purpose, from
First management platform download model, parameter and data.
Further, the system also includes the second management platform, the first default equipment to pass through the second management platform and its
He establishes connection at equipment, and the training of the machine learning model in the first default equipment is sent to other equipment with test data,
Model in other equipment is receiving training with being locally trained, testing after test data, will be local in test passes
Model is transmitted to the first default equipment with parameter by the second management platform, and data include training data and test data.
One or more technical solutions that the application provides, have at least the following technical effects or advantages：
The pace of learning of machine learning model is increased substantially, it is improved and adapts to the ability of different scenes.
Embodiment
The present invention provides a kind of network type machine learning system, and solving existing machine learning, there are pace of learning
Slowly, it is not in contact between each machine, the problem of performance is relatively low, generalization ability is poor, improves the speed of study, can also make same mould
Type is subjected to the inspection of various application scenarios and data so as to improve performance.
To better understand the objects, features and advantages of the present invention, with reference to embodiment
The present invention is further described in detail.It should be noted that in the case where not conflicting mutually, embodiments herein
And the feature in embodiment can be mutually combined.
Many details are elaborated in the following description to facilitate a thorough understanding of the present invention, still, the present invention may be used also
Implemented with the other modes in the range of being different from being described herein using other, therefore, protection scope of the present invention and under
The limitation of specific embodiment disclosed in face.
Traditional machine learning, has and starts from scratch training and carry out two kinds of fine tune based on existing model.Latter side
Formula, is also intended to manually make choice model, then resets training method and data re-start fine tune.
In addition there is a kind of mode for being known as transfer learning, and directly using existing model, then will existing model application
To the technology of frontier.The mode of fine setting is generally also to use existing model, but usually also in original field, typically belongs to one kind
The tune ginseng means of locality.In view of the popularity of artificial intelligence technology future application prospect, it is anticipated that following machine learning
It will be widely used in various scenes, will so have substantial amounts of model and operate in global different location at the same time.If all of
Model all simply carries out offline/on-line training dependent on local data and gradually steps up again, certainly will all make the progress of model
The ability of speed and adaptation scene is greatly lowered.
What machine learning field at present, fine tune and transfer learning were manually carried out, it is limited to each developer's
The factors such as hardware capabilities, data retrieval capabilities, different models are all stand-alone developments.Need by substantial amounts of test ability
Line.Even if on-line running, it is also difficult to tackle all actual scenes completely.
The basic ideas of the present invention allow each machine learning model to exchange " experience " in real time, and " experience " just correspond to certain in fact
Specific machine learning model and its parameter.In this way, making the model for being physically distributed in global different zones, can directly use for reference
The model acquired under other places and scene, and local progress fine will directly be taken and the model got by exchange
Tune carries out transfer learning, so as to improve pace of learning, improves adaptability.
For specific network structure, available mode has several：
Forum type：
I.e. each machine learning model by each self study to model structure and parameter upload on network, while provide scene
With test data.Other machines learning model can directly take when running into similar scene and test data.
Conference type, i.e., for a common theme, carry out online discussion and study by the way of discussion.Such as some
Model test performance under certain scene is inadequate, then online discussion can be initiated in equipment by the model, by scene and test data
It is sent to other models.Other models can it is each it is comfortable local these data are tested, if preferable test result, then
The model of test result and local and parameter can be supplied to meeting promoter, be arrived so as to directly fast and effeciently study ready-made
Model parameter, there is provided the evolutionary rate of machine learning model.
Broadcast type：Existing model and test parameter are collected, again by new model to broadcast after concentration training and after testing
Mode is sent in each equipment in system, then by each equipment replacement model." forum type ", " meeting proposed in the present invention
View formula " " broadcast type " scheme, is only presently preferred embodiments of the present invention, is not intended to limit the invention.It should be pointed out that
All any modification, equivalent and improvement made within the spirit and principles of the invention etc., should be included in the present invention's
Within protection domain.
Although preferred embodiments of the present invention have been described, but those skilled in the art once know basic creation
Property concept, then can make these embodiments other change and modification.So appended claims be intended to be construed to include it is excellent
Select embodiment and fall into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
God and scope.In this way, if these modifications and changes of the present invention belongs to the scope of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to comprising including these modification and variations.
Claims (4)
1. network type machine learning system, it is characterised in that the system comprises：
Each independent machine learning model is run in the equipment of interconnection in system, when in an equipment in system
When model is trained or tests, other equipment provides in the equipment frame of reference model, model parameter and data, to this
The performance of ground model is improved, or in the case of no local model, the model that is directly provided using other models, ginseng
Number, data, establish local model；In system each equipment can be invited or active upload local used in model, parameter, number
According to；The local model of devices in system can actively provide test data, seek to complete to test from system miscellaneous equipment
Model and model parameter.
2. network type machine learning system according to claim 1, it is characterised in that the system also includes the first management
Respective machine learning model structure and parameter is uploaded to the first management platform by platform, each equipment, and to the first management platform
Training and test data are provided；Machine learning on distinct device is according to itself application scenarios and purpose, from the first management platform
Download model, parameter and data.
3. network type machine learning system according to claim 1, it is characterised in that the system also includes the second management
Platform, the machine learning model in the first default equipment is when being tested, by the second management platform in other equipment
Machine learning model establishes connection, and the training of the machine learning model in the first default equipment is sent to other with test data
Machine learning model in equipment, machine learning model in other equipment receive training with after test data in local progress
Training, test, and test result is judged, in test passes, local model and parameter are passed through into the second management platform
It is transmitted to the first default equipment.
4. network type machine learning system according to claim 2, it is characterised in that data include training data and test
Data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711387686.4A CN107977712A (en) | 2017-12-20 | 2017-12-20 | Network type machine learning system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711387686.4A CN107977712A (en) | 2017-12-20 | 2017-12-20 | Network type machine learning system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107977712A true CN107977712A (en) | 2018-05-01 |
Family
ID=62007026
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711387686.4A Pending CN107977712A (en) | 2017-12-20 | 2017-12-20 | Network type machine learning system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107977712A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113449008A (en) * | 2020-03-27 | 2021-09-28 | 华为技术有限公司 | Modeling method and device |
| WO2024092771A1 (en) * | 2022-11-04 | 2024-05-10 | Mediatek Singapore Pte. Ltd. | Model pool at location management function for direct ai/ml positioning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101782976A (en) * | 2010-01-15 | 2010-07-21 | 南京邮电大学 | Automatic selection method for machine learning in cloud computing environment |
| CN104778474A (en) * | 2015-03-23 | 2015-07-15 | 四川九洲电器集团有限责任公司 | Classifier construction method for target detection and target detection method |
| CN106663224A (en) * | 2014-06-30 | 2017-05-10 | 亚马逊科技公司 | Interactive interfaces for machine learning model evaluations |
| CN106909529A (en) * | 2015-12-22 | 2017-06-30 | 阿里巴巴集团控股有限公司 | A kind of Machine learning tools middleware and machine learning training method |
-
2017
- 2017-12-20 CN CN201711387686.4A patent/CN107977712A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101782976A (en) * | 2010-01-15 | 2010-07-21 | 南京邮电大学 | Automatic selection method for machine learning in cloud computing environment |
| CN106663224A (en) * | 2014-06-30 | 2017-05-10 | 亚马逊科技公司 | Interactive interfaces for machine learning model evaluations |
| CN104778474A (en) * | 2015-03-23 | 2015-07-15 | 四川九洲电器集团有限责任公司 | Classifier construction method for target detection and target detection method |
| CN106909529A (en) * | 2015-12-22 | 2017-06-30 | 阿里巴巴集团控股有限公司 | A kind of Machine learning tools middleware and machine learning training method |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113449008A (en) * | 2020-03-27 | 2021-09-28 | 华为技术有限公司 | Modeling method and device |
| WO2021190068A1 (en) * | 2020-03-27 | 2021-09-30 | 华为技术有限公司 | Model building method and device |
| CN113449008B (en) * | 2020-03-27 | 2023-06-06 | 华为技术有限公司 | Modeling method and device |
| WO2024092771A1 (en) * | 2022-11-04 | 2024-05-10 | Mediatek Singapore Pte. Ltd. | Model pool at location management function for direct ai/ml positioning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102195922B1 (en) | Internet-based facial beautification system | |
| CN109387204A (en) | The synchronous positioning of the mobile robot of dynamic environment and patterning process in faced chamber | |
| CN109635748B (en) | Method for extracting road characteristics in high-resolution image | |
| CN107977712A (en) | Network type machine learning system | |
| CN108686978A (en) | The method for sorting and system of fruit classification and color and luster based on ARM | |
| CN112001709A (en) | Intelligent assignment scheduling AR multi-person remote assistance method based on ant colony algorithm | |
| Ou et al. | A novel rank learning based no-reference image quality assessment method | |
| CN108009497A (en) | Image recognition monitoring method, system, computing device and readable storage medium storing program for executing | |
| CN114092769A (en) | Transformer substation multi-scene inspection analysis method based on federal learning | |
| CN109522775A (en) | Face character detection method, device and electronic equipment | |
| CN110223251A (en) | Suitable for manually with the convolutional neural networks underwater image restoration method of lamp | |
| CN107845129A (en) | Three-dimensional reconstruction method and device, the method and device of augmented reality | |
| CN110288878A (en) | Adaptive learning method and device | |
| CN110070059A (en) | A kind of unstructured road detection method based on domain migration | |
| CN108891421A (en) | A method of building driving strategy | |
| CN113241151B (en) | Intelligent radiotherapy cloud system and implementation method | |
| CN116433470A (en) | Model training method, data enhancement method, target detection method and related equipment | |
| Wan | The Application Strategy of Intelligent Computer Multimedia Technology in Film and Television Post Production | |
| CN114529484A (en) | Deep learning sample enhancement method for direct current component change in imaging | |
| CN111950416B (en) | Face recognition method and system based on block chain | |
| Wang et al. | An improved reinforcement learning method for underwater image enhancement | |
| CN107509083B (en) | Live streaming number authenticity detection method, storage medium, electronic equipment and system | |
| CN106250313A (en) | Mobile phone terminal VoLte video interconnection automated testing method and device | |
| CN107659805A (en) | A kind of video conversion method and device | |
| CN112926088B (en) | Federal learning privacy policy selection method based on game theory |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180501 |
|
| RJ01 | Rejection of invention patent application after publication |