CN112966601A - Method for artificial intelligence teachers and apprentices to learn by semi-supervision - Google Patents
Method for artificial intelligence teachers and apprentices to learn by semi-supervision Download PDFInfo
- Publication number
- CN112966601A CN112966601A CN202110244277.9A CN202110244277A CN112966601A CN 112966601 A CN112966601 A CN 112966601A CN 202110244277 A CN202110244277 A CN 202110244277A CN 112966601 A CN112966601 A CN 112966601A
- Authority
- CN
- China
- Prior art keywords
- data
- identification
- supervision
- adjustment
- model
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The invention discloses a method for artificial intelligence teachers and apprentices to learn by semi-supervision, and relates to the technical field of AI model algorithms. The method for the artificial intelligence teacher and apprentice semi-supervised learning comprises the following steps: analyzing and processing the identification data, processing the independent data, judging whether the identification data needs high-precision model supervision and adjustment, judging whether the identification data needs third-party algorithm supervision and adjustment, and automatically adjusting the data set by a background management interface, an application interface of a front-end user and a set time node through manual intervention supervision and adjustment by the user. The method for the artificial intelligence teacher-apprentice semi-supervised learning formulates a set of effective teacher-apprentice semi-supervised learning method, thereby realizing repeated retraining of the AI model in an actual scene with the lowest cost. And the method continuously adapts to new actual scenes, and simultaneously further optimizes the training data set, thereby finally achieving the purpose of greatly improving the accuracy of the model.
Description
Technical Field
The invention relates to the technical field of AI model algorithms, in particular to a method for artificial intelligence teachers and apprentices to semi-supervised learning.
Background
The models of the common visual artificial intelligence bodies are trained. The training data is a static data set, meaning that the accuracy of the visual artificial intelligence model is given as the training of the model is completed. In actual use, there may be a large difference between the real scene and the training data. For example, the deviation of the video camera angle, the deviation of the distance, the deviation of day and night, etc. cause problems for the accurate judgment of the model. Object identification and employee wearing standard identification also can be different along with the change of time, for example, cell-phone change between ten years can lead to cell-phone identification using old data, and the accuracy is not high. The method of manually updating a machine learning model is essentially a process of replicating initial training data, but using an updated set of data inputs. The feasibility of this option depends on the ability to periodically acquire and prepare new training data. The advantage is that the performance of the model can be monitored at any time to determine when an update is required. If the accuracy of the model drops significantly, the updated data may need to be retrained. The biggest problem, the labor cost associated with data preparation, and the uncertainty associated with retraining.
Continuous learning models typically incorporate new data streams from the production environment in which the data is deployed. After the user uses the existing machine learning model, the daily recognition result can form recognition records, alarm records and the like. During the use process of the data, the feedback of the user can be obtained, and new identification data is formed by the data. Human intervention is possible in the continuous learning model, but it represents a real bottleneck.
The first problem of the above technology is that an effective and feasible method for continuous retraining by automatically using the recognition data of the actual scene is not provided, the second problem is that the ability of further intelligentizing the human intervention is not provided, and the third problem is that the accuracy difference of the machine training model under different hardware is ignored.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides a method for artificial intelligence teachers and apprentices to learn by semi-supervision, and solves the problems mentioned in the background technology.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme: the utility model provides a method of artificial intelligence teachers and apprentices semi-supervised learning, includes the identification data under the in-service use scene, and identification data adds and replaces original identification model after, becomes new model data, and identification data has included the data adjustment when the model is synthesized, has contained user's artificial intervention supervision adjustment and third party algorithm supervision adjustment simultaneously, carries out the secondary adjustment to type data in appointed cycle, has formed an automatic training flow, artificial intelligence teachers and apprentices semi-supervised learning's method includes following step:
and S1, analyzing and processing the identification data, recording an original video picture, an identification frame, identification confidence coefficient and the like after the identification data are generated in the actual operation process, and uploading the identification data to a cloud end through a network for storage.
And S2, independent data processing flow, wherein the processing flow is executed only on the actual scene identification data appointed in the background management interface, after the identification data enter, the processing flow is started, the integrated identification supervision adjustment data is supplemented, the result is operated, and the state data is added in the integrated identification supervision adjustment column of the identification data.
And S3, judging whether the identification data needs high-precision model supervision and adjustment, supplementing the high-precision model supervision and adjustment data when the threshold value or the set condition of management configuration is met, then operating the result, identifying the high-precision model supervision and adjustment column of the data, and adding state data.
And S4, judging whether the identification data needs to be supervised and adjusted by a third-party algorithm, supplementing the supervised and adjusted data by the third-party algorithm under the condition of meeting a threshold or a set management configuration, operating the result, identifying the supervised and adjusted column of the third-party algorithm of the data, and adding state data.
And S5, manually intervening, supervising and adjusting the user, inputting the user feedback into the identification data in an asynchronous mode through a background management interface and an application interface of a front-end user, confirming the user feedback according to the actual feedback condition by the cloud artificial intelligence training personnel, updating the thought intervening, supervising and adjusting column of the user, adjusting the data of S2 according to the actual condition, and starting an execution process from S2.
And S6, automatically adjusting the data set according to the set time node, and starting the algorithm training process of the model data.
Preferably, in S2, the comprehensive identification supervision adjustment column performs identification on multiple data in an actual scene, and performs judgment on comprehensive aspects by using multiple models, and the identification results are summarized and accumulated.
Preferably, in S3, the high-precision model is supervised and adjusted in an actual scene, and an edge-side computing device is used for detection, so that the used identification model is a small low-precision model on the premise of ensuring the size of a hardware memory and the normal performance of the field identification sensitivity, and under such an identification condition, secondary verification is performed on the identification data and the cloud high-precision large model.
Preferably, different algorithms are adopted for supervision and adjustment of the third-party algorithm in S4, accuracy is higher in a specific scene, and by using the difference of the algorithms, recognition speed can be distinguished, so that some algorithms with better recognition degree cannot be applied in an actual scene, and the algorithms can be automatically supervised on an actual recognition result in a cloud.
Preferably, in S5, the user performs human intervention to supervise and adjust, and in the process of actually recognizing data, obtains actual data feedback of the user, and performs background manual processing and induction on the feedback data information, and performs data verification again, and the feedback data information becomes an important label for retraining human intervention.
Preferably, in S6, the data set is automatically changed and adjusted, and the result of the algorithm training of the model data is subjected to re-inspection of the historical high-confidence data after training to evaluate the degree of evolution of the entire model data, and the addition and change ratios of the training data set, so that the training code is automatically changed and adjusted according to the degree of evolution.
Preferably, the identification data in S1 is uploaded to the cloud for storage through a network, and the network transmission mode adopted is one or more of wired network transmission, routed wireless network transmission, and 5G network transmission.
Preferably, the model data is in a constantly changing state, and the high-confidence data is compared and fused all the time.
Preferably, the information data content of the identification data is not completely the same, and the degree of fusion with the model data is not completely the same.
(III) advantageous effects
The invention provides a method for artificial intelligence teachers and apprentices to learn by semi-supervision. The method has the following beneficial effects:
(1) the method for the artificial intelligence teachers and apprentices semi-supervised learning formulates an effective teachers and apprentices semi-supervised learning method, so that repeated retraining of the AI model under an actual scene is realized with the lowest cost. And the method continuously adapts to new actual scenes, and simultaneously further optimizes the training data set, thereby finally achieving the purpose of greatly improving the accuracy of the model.
(2) The method for the artificial intelligence teacher-apprentice semi-supervised learning defines different automatic and artificial supervision types, provides methods of comprehensive identification supervision adjustment, high-precision model supervision adjustment and third-party algorithm supervision adjustment, and explains a data supplement process of a teacher-apprentice semi-supervised learning process on a cloud server. In addition, a method for retraining through automatic adjustment of the data set after the completion of the supplement of the identification data is also provided.
Drawings
FIG. 1 is a schematic view of a flow structure according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, the present invention provides a technical solution: the utility model provides a method of artificial intelligence teachers and apprentices semi-supervised learning, including the identification data under the in-service use scene, after the identification data adds and replaces original identification model, become new model data, the identification data has included the data adjustment when the model is synthesized, contained user's artificial intervention supervision adjustment and third party algorithm supervision adjustment simultaneously, carry out the secondary adjustment to type data in appointed cycle, formed an automatic training flow, identification data in the implementation of this year represents teachers and fathers ' experience, and the signal that the model data representation apprentices and accept, artificial intelligence teachers and apprentices semi-supervised learning's method includes following step:
and S1, analyzing and processing the identification data, recording an original video picture, an identification frame, identification confidence coefficient and the like after the identification data are generated in the actual operation process, and uploading the identification data to a cloud end through a network for storage.
And S2, independent data processing flow, wherein the processing flow is executed only on the actual scene identification data appointed in the background management interface, after the identification data enter, the processing flow is started, the integrated identification supervision adjustment data is supplemented, the result is operated, and the state data is added in the integrated identification supervision adjustment column of the identification data.
And S3, judging whether the identification data needs high-precision model supervision and adjustment, supplementing the high-precision model supervision and adjustment data when the threshold value or the set condition of management configuration is met, then operating the result, identifying the high-precision model supervision and adjustment column of the data, and adding state data.
And S4, judging whether the identification data needs to be supervised and adjusted by a third-party algorithm, supplementing the supervised and adjusted data by the third-party algorithm under the condition of meeting a threshold or a set management configuration, operating the result, identifying the supervised and adjusted column of the third-party algorithm of the data, and adding state data.
And S5, manually intervening, supervising and adjusting the user, inputting the user feedback into the identification data in an asynchronous mode through a background management interface and an application interface of a front-end user, confirming the user feedback according to the actual feedback condition by the cloud artificial intelligence training personnel, updating the thought intervening, supervising and adjusting column of the user, adjusting the data of S2 according to the actual condition, and starting an execution process from S2.
And S6, automatically adjusting the data set according to the set time node, and starting the algorithm training process of the model data.
Further, in this embodiment, in S2, the comprehensive identification supervision adjustment column performs identification on multiple data in an actual scene, and performs judgment on the comprehensive aspect by using multiple models, and the identification results are summarized and accumulated.
Further, in this embodiment, in S3, the high-precision model is supervised and adjusted in an actual scene, and an edge-side computing device is used for detection, so that the used identification model is a small low-precision model on the premise of ensuring that the size of the hardware memory and the field identification sensitivity can be normally exerted, and under such an identification condition, secondary verification is performed on the identification data and the cloud high-precision large model.
Further, in this embodiment, different algorithms are adopted for the third-party algorithm supervision and adjustment in S4, the accuracy is higher in a specific scene, and by using the difference of the algorithms, the recognition speed can be distinguished, so that some algorithms with better recognition degree cannot be applied in an actual scene, and the algorithms can be automatically supervised on an actual recognition result in a cloud.
Further, in this embodiment, in S5, the user performs human intervention and supervision adjustment, and in the process of actually recognizing data, actual data feedback of the user is obtained, and the feedback data information is subjected to background manual processing and summarization, and is verified again, and becomes an important label for retraining manual intervention.
Further, in this embodiment, in S6, the data set is automatically changed and adjusted, and as a result of the algorithm training of the model data, the degree of evolution of the entire model data, the addition and change ratio of the training data set are evaluated by rechecking the historical high-confidence data after training, and the training code is automatically changed and adjusted according to the degree of evolution.
Further, in this embodiment, the identification data in S1 is uploaded to a cloud for storage through a network, and the network transmission mode adopted is one or more of wired network transmission, routed wireless network transmission, and 5G network transmission.
Further, in this embodiment, the model data is in a constantly changing state, and the comparison and fusion are performed on the high confidence data all the time.
Further, in this embodiment, the information data content of the identification data is not completely the same, and the degree of fusion with the model data is not completely the same.
The method for the artificial intelligence teacher-apprentice semi-supervised learning formulates a set of effective teacher-apprentice semi-supervised learning method, thereby realizing repeated retraining of the AI model in an actual scene with the lowest cost. And the method continuously adapts to new actual scenes, and simultaneously further optimizes the training data set, thereby finally achieving the purpose of greatly improving the accuracy of the model. The method for comprehensive identification, supervision and adjustment, high-precision model supervision and adjustment and third-party algorithm supervision and adjustment is provided, and a data supplement process of a teacher-apprentice semi-supervised learning process on a cloud server is described. In addition, a method for retraining through automatic adjustment of the data set after the completion of the supplement of the identification data is also provided.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (9)
1. A method for artificial intelligence teachers and apprentices to learn by semi-supervision is characterized in that: including the identification data under the actual use scene, after the identification data adds and replaces original identification model, become new model data, the data adjustment when the identification data has included the model synthesis has contained user's artificial intervention supervision adjustment and third party algorithm supervision adjustment simultaneously, carries out the secondary adjustment to type data in appointed cycle, has formed an automatic training flow, the method of artificial intelligence teachers and apprentices semi-supervised learning includes following steps:
s1, analyzing and processing the identification data, recording an original video picture, an identification frame, identification confidence coefficient and the like after the identification data are generated in the actual operation process, and uploading the identification data to a cloud end through a network for storage;
s2, independent data processing flow, wherein the processing flow is executed only on the designated actual scene identification data in the background management interface, after the identification data enters, the processing flow is started, the integrated identification supervision adjustment data starts to be supplemented, the result is operated, and the state data is added in the integrated identification supervision adjustment column of the identification data;
s3, judging whether the identification data needs high-precision model supervision and adjustment, supplementing the high-precision model supervision and adjustment data when the threshold value or the set condition of management configuration is met, then operating the result, identifying the high-precision model supervision and adjustment column of the data, and adding state data;
s4, judging whether the identification data needs third-party algorithm supervision and adjustment, supplementing the third-party algorithm supervision and adjustment data when the threshold value or the set condition of management configuration is met, then operating the result, identifying the third-party algorithm supervision and adjustment column of the data, and adding state data;
s5, carrying out human intervention supervision and adjustment on the user through a background management interface and an application interface of a front-end user, inputting the user feedback into identification data in an asynchronous mode, confirming the user feedback by a cloud artificial intelligence trainer according to the actual feedback condition, updating a user thought intervention supervision adjustment column, carrying out data adjustment on S2 according to the actual condition, and starting an execution flow from S2;
and S6, automatically adjusting the data set according to the set time node, and starting the algorithm training process of the model data.
2. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: in the S2, the comprehensive identification supervision adjustment column performs identification on multiple data in an actual scene, and adopts multiple models to perform judgment on the comprehensive aspect, and the identification results are summarized and accumulated.
3. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: in the S3, the high-precision model is supervised and adjusted in an actual scene, edge-side computing equipment is used for detection, the used identification model is a small low-precision model on the premise that the size of a hardware memory and the identification sensitivity of a field can be normally exerted, and under the identification condition, secondary verification is carried out on identification data and a cloud high-precision large model.
4. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: in the S4, different algorithms are adopted for the third-party algorithm supervision and adjustment, the accuracy is higher in a specific scene, the recognition speed can be distinguished by using the difference of the algorithms, so that algorithms with better recognition degree cannot be applied to the actual scene, and the algorithms can be automatically supervised on the actual recognition result at the cloud.
5. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: s5, the user intervenes and supervises and adjusts manually, in the process of identifying data actually, the actual data feedback of the user is obtained, the feedback data information is processed and summarized manually in the background, the data is verified again, and the feedback data information becomes an important mark of retraining manual intervention.
6. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: and automatically changing and adjusting the data set in S6, and automatically changing and adjusting the evolution degree of the whole model data, the addition and change proportion of the training data set and the learned training code according to the evolution degree by rechecking the historical high-confidence data after training the algorithm training result of the model data.
7. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: and in the S1, the identification data is uploaded to a cloud end through a network for storage, and the adopted network transmission mode is one or more of wired network transmission, routing wireless network transmission and 5G network transmission.
8. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: the model data is in a constantly changing state, and high-confidence data is compared and fused all the time.
9. The method of claim 1 for artificial intelligence teachers and apprentices semi-supervised learning, which is characterized in that: the information data content of the identification data is not completely the same, and the degree of fusion with the model data is not completely the same.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110244277.9A CN112966601A (en) | 2021-03-05 | 2021-03-05 | Method for artificial intelligence teachers and apprentices to learn by semi-supervision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110244277.9A CN112966601A (en) | 2021-03-05 | 2021-03-05 | Method for artificial intelligence teachers and apprentices to learn by semi-supervision |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112966601A true CN112966601A (en) | 2021-06-15 |
Family
ID=76276762
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110244277.9A Pending CN112966601A (en) | 2021-03-05 | 2021-03-05 | Method for artificial intelligence teachers and apprentices to learn by semi-supervision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112966601A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537168A (en) * | 2018-04-09 | 2018-09-14 | 云南大学 | Human facial expression recognition method based on transfer learning technology |
| CN109949317A (en) * | 2019-03-06 | 2019-06-28 | 东南大学 | Based on the semi-supervised image instance dividing method for gradually fighting study |
| CN110462591A (en) * | 2017-04-26 | 2019-11-15 | 美的集团股份有限公司 | Using Job Server on large scale distributed system training machine learning model |
| CN110705684A (en) * | 2019-08-22 | 2020-01-17 | 中国科学院计算技术研究所 | Environment self-adaptive learning method and system based on end cloud cooperation |
| CN110705630A (en) * | 2019-09-27 | 2020-01-17 | 聚时科技(上海)有限公司 | Semi-supervised learning type target detection neural network training method, device and application |
| CN111368660A (en) * | 2020-02-25 | 2020-07-03 | 华南理工大学 | Single-stage semi-supervised image human body target detection method |
| CN112417986A (en) * | 2020-10-30 | 2021-02-26 | 四川天翼网络服务有限公司 | Semi-supervised online face recognition method and system based on deep neural network model |
-
2021
- 2021-03-05 CN CN202110244277.9A patent/CN112966601A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110462591A (en) * | 2017-04-26 | 2019-11-15 | 美的集团股份有限公司 | Using Job Server on large scale distributed system training machine learning model |
| CN108537168A (en) * | 2018-04-09 | 2018-09-14 | 云南大学 | Human facial expression recognition method based on transfer learning technology |
| CN109949317A (en) * | 2019-03-06 | 2019-06-28 | 东南大学 | Based on the semi-supervised image instance dividing method for gradually fighting study |
| CN110705684A (en) * | 2019-08-22 | 2020-01-17 | 中国科学院计算技术研究所 | Environment self-adaptive learning method and system based on end cloud cooperation |
| CN110705630A (en) * | 2019-09-27 | 2020-01-17 | 聚时科技(上海)有限公司 | Semi-supervised learning type target detection neural network training method, device and application |
| CN111368660A (en) * | 2020-02-25 | 2020-07-03 | 华南理工大学 | Single-stage semi-supervised image human body target detection method |
| CN112417986A (en) * | 2020-10-30 | 2021-02-26 | 四川天翼网络服务有限公司 | Semi-supervised online face recognition method and system based on deep neural network model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111353413B (en) | Low-missing-report-rate defect identification method for power transmission equipment | |
| CN110443969B (en) | Fire detection method and device, electronic equipment and storage medium | |
| CN107680195A (en) | A kind of transformer station intelligent robot inspection Computer Aided Analysis System and method | |
| CN110807353A (en) | Transformer substation foreign matter identification method, device and system based on deep learning | |
| CN105426908B (en) | A kind of substation's attributive classification method based on convolutional neural networks | |
| CN107305636A (en) | Target identification method, Target Identification Unit, terminal device and target identification system | |
| CN111832400B (en) | Mask wearing condition monitoring system and method based on probabilistic neural network | |
| CN106845432A (en) | The method and apparatus that a kind of face is detected jointly with human body | |
| CN107194396A (en) | Method for early warning is recognized based on the specific architecture against regulations in land resources video monitoring system | |
| CN114359727A (en) | Tea disease identification method and system based on lightweight optimization Yolo v4 | |
| CN109754089A (en) | A kind of model training systems and method | |
| CN116052222A (en) | Cattle face recognition method for naturally collecting cattle face image | |
| CN115661650A (en) | Farm management system based on data monitoring of Internet of things | |
| CN112396042A (en) | Real-time updated target detection method and system, and computer-readable storage medium | |
| CN111651890B (en) | Data-driven aluminum electrolysis digital twin factory, control method and system | |
| CN117215209A (en) | Cloud home control method and system based on Internet of things | |
| CN112966601A (en) | Method for artificial intelligence teachers and apprentices to learn by semi-supervision | |
| CN116884192A (en) | Power production operation risk early warning method, system and equipment | |
| CN116824369A (en) | Litchi diseases real-time detection system and method based on edge calculation and yolov7-tiny | |
| CN113610191B (en) | Garbage classification model modeling method and garbage classification method | |
| CN117808374B (en) | Intelligent acceptance management method and system for building engineering quality | |
| CN111931598B (en) | Intelligent real-time classroom analysis method and system based on face recognition | |
| CN111970489B (en) | Intelligent monitoring management method and system based on man-machine two-way | |
| CN114155555B (en) | Human behavior artificial intelligence judgment system and method | |
| CN112396033A (en) | Bird background rhythm detection method and device, terminal equipment and storage medium |
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: 20210615 |
|
| RJ01 | Rejection of invention patent application after publication |