CN107247972A - One kind is based on mass-rent technology classification model training method - Google Patents
One kind is based on mass-rent technology classification model training method Download PDFInfo
- Publication number
- CN107247972A CN107247972A CN201710511119.9A CN201710511119A CN107247972A CN 107247972 A CN107247972 A CN 107247972A CN 201710511119 A CN201710511119 A CN 201710511119A CN 107247972 A CN107247972 A CN 107247972A
- Authority
- CN
- China
- Prior art keywords
- mrow
- sample
- training
- msubsup
- markup information
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Mass-rent technology classification model training method is based on the invention provides one kind.The level that user provides markup information is estimated on the corresponding mass-rent markup information of a small amount of sample;Markup information used in training sample is determined by the use of observed mark person level as priori;The train classification models on training sample and its markup information;The training sample for using the disaggregated model selecting that the model anticipation error can be made minimum, and predict the sample generic;Training set is added by selected sample and in the flat highest user of category subscript water filling for its markup information provided;On training set in the updated, iteration performs above-mentioned steps, untill the precision of disaggregated model or the quantity of training sample reach preset standard.Effect of the invention is that, it is to avoid adverse effect of the low quality markup information that the low user of mark level provides to disaggregated model training, it is ensured that train the effect of high generalization ability disaggregated model under mass-rent environment.
Description
Technical field
The present invention relates to a kind of disaggregated model training method.
Background technology
At present, in machine learning under the framework of supervised learning, train classification models need to collect one group in advance with mark letter
The data sample of breath.The quality and quantity of collected training data directly determines the Generalization Capability of disaggregated model.Traditional
Training data provides data sample corresponding unique correct mark letter during collecting it is necessary to have the expert of professional domain knowledge
Breath, for ensureing that the disaggregated model obtained by training has good Generalization Capability.
This traditional method facing challenges is that the personnel with specialty background are less in realistic task, obtains sample mark
The cost of information is higher, the time is longer.Thus, with the development of network technology and data storage technology, it is using mass-rent technology
The a large amount of cheap markup informations of training sample quick obtaining, it is effective as the time and economic cost reduced in mark acquisition process
One of approach.
Under mass-rent environment, the mark acquisition task of training data is not completed by traditional professional, but with freedom certainly
The form of hope is contracted out to unspecific popular network to complete, i.e., amateur personal or individual of increasing income is with independent or cooperation side
The quick qurer of formula completes mark task.Because the markup information based on mass-rent technical limit spacing is used from multiple online networks
Family, therefore, it is difficult to ensure the quality of collected markup information, simultaneously as the correct markup information for lacking professional person's offer is made
For " goldstandard ", it is also difficult to which the degree of accuracy of experience and its completion mark task to these users is weighed.Directly using many
Bag markup information train classification models, can have a strong impact on the Generalization Capability of disaggregated model.
The content of the invention
It is an object of the invention to provide a kind of influence that can overcome low quality markup information to model training process, it is ensured that
With the disaggregated model of minimum labeled cost one high generalization ability of study based on mass-rent technology classification model under mass-rent environment
Training method.
The object of the present invention is achieved like this:
It is in m collected sample and by the mass-rent markup information of k user's offerCondition
Under, carry out in accordance with the following steps:
Step one, n sample and its corresponding mass-rent mark are randomly selected from collected sample and its mass-rent labeled data
Information
Step 2, builds training datasetWherein, whenWhen, yi=1, otherwise, yi=0;
Step 3, in training datasetThe disaggregated model that upper one parameter of study is w;
Step 4, one group of markup information that j-th of user provides on classification cOn mark level
For
Wherein,WithRepresent that the user provides the number of times of correct mark and mistake mark respectively;
Step 5, according to for sample xiThe mark level of multiple users of markup information is provided, the sample is used for the mark trained
Information is estimated by following formula
Step 5, is predicted using disaggregated model to remaining m-n sample generic, and calculates corresponding point of each sample
Class model anticipation error, it is as follows
Wherein, U represents the set of remaining sample composition, and (D x) represents to add sample into the error of disaggregated model after training set I;
Step 6, and selection p (y | x*;W) the corresponding classifications of > 0.5, and be sample by the flat highest user of category subscript water filling
x*The markup information y of offer*Add training dataset
Step 7, repeats step 3 to step 6, until the Generalization accuracy of disaggregated model or the quantity of training sample reach
Untill written standards.
The present invention can also include:
1st, mark level of j-th of user on classification cComputational methods be:
In training sample and its mass-rent mark setOn, initialization sample xiCorresponding mark letter
Cease yi, that is, work asWhen, yi=1, otherwise, yi=0;According to markup information yi, initialize the mark level of multiple users
Collect θ, as prior distribution, markup information corresponding to training data is estimated as follows:
Wherein
p(yi|xi;W) it is probabilistic estimated value of the disaggregated model to training sample generic, according to estimated
Update user annotation level, and train classification models again;Repeat to estimateWith update user annotation level, until seemingly
Right functionConvergence, this process terminates.
2nd, sample is added into the error I of disaggregated model after training set (D, computational methods x) is:
Wherein, U' represent by the sample from remaining sample set deletion after sample set;Meanwhile, when sample x adds markup information
And add after training set, the disaggregated model obtained by study is to the prediction probability of the samplexu∈
U ', yu∈{0,1}。
The present invention proposes a kind of based on mass-rent technology classification model training method.In the markup information provided from unique user
Estimate the mark level of the user, the quality of markup information is provided according to multiple users come used in determining train classification models
Markup information, reduction low-level user provides negative effect of the low quality markup information to disaggregated model training.Selected section sample
This structure training dataset, improves the Generalization Capability of disaggregated model, it is ensured that the application effect of disaggregated model in actual task.
The purpose of the present invention is multiple low quality markup informations using mass-rent technical limit spacing, with minimum labeled cost study one
The disaggregated model of individual high generalization ability.Beneficial effects of the present invention:There is provided present invention utilizes mass-rent user for training sample
Multiple markup informations estimate that unique user provides the level of mark in each classification, using the mark level of multiple users as
Priori, the sample markup information that estimation train classification models are used, overcomes low quality markup information to model training mistake
The influence of journey.According to class prediction result of the disaggregated model to remaining sample, selection makes the minimum sample of disaggregated model anticipation error
This addition training sample set, meanwhile, the markup information that the flat highest user of selection selection category subscript water filling provides adds instruction
Practice markup information collection, overcome interference of the low quality sample to disaggregated model training.So as to, it is ensured that with most under mass-rent environment
Small labeled cost learns the disaggregated model of a high generalization ability.
Brief description of the drawings
Fig. 1 is flow chart of the invention;
Fig. 2 is the present invention and the precision comparative result using correct mark train classification models;
Fig. 3 for the true mark level of the user annotation level estimated of the present invention and user comparative result.
Embodiment
Illustrate below and the present invention is described in more detail.
According to the flow chart based on mass-rent technology classification model training process in the present invention, comprise the following steps that:
1) n (n < m) individual samples and corresponding mass-rent markup information are randomly selected from collected data
2) training sample x is determinediCorresponding markup information yi, whenWhen, yi=1, otherwise, yi=0.
3) disaggregated model that a parameter is w is being learnt using training sample and its markup information.
4) in training sample and its mass-rent mark setOn, according to markup information yi, estimate multiple
The corresponding mark level set θ of user, wherein, give one group of markup information that j-th of user provides on classification cThe user marks level accordinglyFor:
Wherein,WithRepresent that the user provides the number of times of correct mark and mistake mark respectively.As prior distribution,
Markup information corresponding to training data is estimated as follows:
Wherein:
p(yi|xi;W) it is probabilistic estimated value of the disaggregated model to training sample generic.According to estimated
Update user annotation level, and train classification models again.Repeat to estimateWith update user annotation level, until seemingly
Right functionConvergence, this process terminates.
5) basis is sample xiThere is provided markup information multiple users mark level, update the sample be used for train mark believe
Breath:
6) remaining m-n sample generic is predicted according to using disaggregated model, and calculates the corresponding of each sample
Disaggregated model anticipation error, it is as follows:
Wherein:
U represents the set of remaining sample composition, and U' is represented the sample from the sample set after the deletion of remaining sample set.Meanwhile,
After sample x adds markup information and adds training set, the disaggregated model obtained by study is to the prediction probability of the samplexu∈ U ', yu∈{0,1}。
7) selection p (y | x*;W) the corresponding classifications of > 0.5, and be sample x by the flat highest user of category subscript water filling*There is provided
Markup information y*Add training dataset
8) repeat step 3 to step 6, until the Generalization accuracy of disaggregated model or the quantity of training sample reach it is set
Untill standard.
Because in the training process of disaggregated model, the mass-rent markup information of selected sample is only put down by category subscript water filling
Highest mark person is provided, and then the level for the person that use mark is corrected and marked used in training sample as priori
Information, it is ensured that the quality of markup information is used during train classification models, reduces low-level mark person and provides mark letter
Cease the negative effect to disaggregated model training process.Secondly, in disaggregated model training process, for disaggregated model to choose
With training sample, it is ensured that disaggregated model can make full use of the information that sample is included.
In the simulation process to this algorithm, 5 users are used altogether to simulate mass-rent markup information, training dataset bag
40 training samples are contained, unlabeled data concentration contains 1000 and do not mark sample, and test data is concentrated and contains 1000
Bar sample.Fig. 2 is shown under the conditions of proposed mass-rent mark and distinguished under the conditions of correct mark the ROC essences of learning classification model
Spend comparative result.Fig. 3 is shown proposed method and estimates that obtained user's level and user are true according to mass-rent markup information
The comparative result of real level.
Although having been incorporated with embodiment to enter one kind of the present invention based on mass-rent technology classification model training method
Go explanation, but the invention is not restricted to this.The various modifications made under spirit and principles of the present invention should be included in this
Within the scope of claims of invention are limited.
Claims (3)
1. one kind is based on mass-rent technology classification model training method, it is characterized in that in m collected sample and by k use
Family provide mass-rent markup information beUnder conditions of, carry out in accordance with the following steps:
Step one, n sample and its corresponding mass-rent mark are randomly selected from collected sample and its mass-rent labeled data
Information
Step 2, builds training datasetWherein, whenWhen, yi=1, otherwise, yi=0;
Step 3, in training datasetThe disaggregated model that upper one parameter of study is w;
Step 4, one group of markup information that j-th of user provides on classification cOn mark levelFor
<mrow>
<msubsup>
<mi>&theta;</mi>
<mi>j</mi>
<mi>c</mi>
</msubsup>
<mo>=</mo>
<mi>B</mi>
<mi>e</mi>
<mi>t</mi>
<mi>a</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&theta;</mi>
<mi>j</mi>
<mi>c</mi>
</msubsup>
<mo>|</mo>
<msubsup>
<mi>&alpha;</mi>
<mi>j</mi>
<mn>1</mn>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>&alpha;</mi>
<mi>j</mi>
<mn>2</mn>
</msubsup>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>c</mi>
<mo>&Element;</mo>
<mo>{</mo>
<mn>0</mn>
<mo>,</mo>
<mn>1</mn>
<mo>}</mo>
</mrow>
Wherein,WithRepresent that the user provides the number of times of correct mark and mistake mark respectively;
Step 5, according to for sample xiThe mark level of multiple users of markup information is provided, the sample is used for the mark letter trained
Breath is estimated by following formula
<mrow>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<munder>
<mi>argmax</mi>
<mrow>
<mi>y</mi>
<mo>&Element;</mo>
<mo>{</mo>
<mn>0</mn>
<mo>,</mo>
<mn>1</mn>
<mo>}</mo>
</mrow>
</munder>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>y</mi>
<mo>|</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>;</mo>
<mi>w</mi>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<munderover>
<mo>&Pi;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>k</mi>
</munderover>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>y</mi>
<mi>i</mi>
<mi>j</mi>
</msubsup>
<mo>|</mo>
<mi>y</mi>
<mo>;</mo>
<msub>
<mi>&theta;</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Step 5, is predicted using disaggregated model to remaining m-n sample generic, and calculates corresponding point of each sample
Class model anticipation error, it is as follows
<mrow>
<msup>
<mi>x</mi>
<mo>*</mo>
</msup>
<mo>=</mo>
<munder>
<mrow>
<mi>arg</mi>
<mi>min</mi>
</mrow>
<mrow>
<mi>x</mi>
<mo>&Element;</mo>
<mi>U</mi>
</mrow>
</munder>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>y</mi>
<mo>&Element;</mo>
<mo>{</mo>
<mn>0</mn>
<mo>,</mo>
<mn>1</mn>
<mo>}</mo>
</mrow>
</munder>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>y</mi>
<mo>|</mo>
<mi>x</mi>
<mo>;</mo>
<mi>w</mi>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>,</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mrow>
Wherein, U represents the set of remaining sample composition, and (D x) represents to add sample into the error of disaggregated model after training set I;
Step 6, and selection p (y | x*;W) the corresponding classifications of > 0.5, and be sample x by the flat highest user of category subscript water filling*
The markup information y of offer*Add training dataset
Step 7, repeats step 3 to step 6, until the Generalization accuracy of disaggregated model or the quantity of training sample reach
Untill written standards.
2. according to claim 1 be based on mass-rent technology classification model training method, it is characterized in that j-th of user is in class
Mark level on other cComputational methods be:
In training sample and its mass-rent mark setOn, initialization sample xiCorresponding mark letter
Cease yi, that is, work asWhen, yi=1, otherwise, yi=0;According to markup information yi, initialize the mark water of multiple users
Flat collection θ, as prior distribution, markup information corresponding to training data is estimated as follows:
Wherein
<mrow>
<mi>C</mi>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>&Element;</mo>
<mo>{</mo>
<mn>0</mn>
<mo>,</mo>
<mn>1</mn>
<mo>}</mo>
</mrow>
</munder>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>;</mo>
<mi>w</mi>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<munderover>
<mo>&Pi;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>k</mi>
</munderover>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>y</mi>
<mi>i</mi>
<mi>j</mi>
</msubsup>
<mo>|</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>;</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
</mrow>
p(yi|xi;W) it is probabilistic estimated value of the disaggregated model to training sample generic, according to estimatedMore
New user annotation level, and train classification models again;Repeat to estimateWith update user annotation level, until likelihood
FunctionConvergence, this process terminates.
3. according to claim 1 or 2 be based on mass-rent technology classification model training method, it is characterized in that sample is added
(D, computational methods x) are the error I of disaggregated model after training set:
<mrow>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>,</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>u</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mo>|</mo>
<msup>
<mi>U</mi>
<mo>&prime;</mo>
</msup>
<mo>|</mo>
</mrow>
</munderover>
<munder>
<mo>&Sigma;</mo>
<mi>k</mi>
</munder>
<msubsup>
<mi>p</mi>
<mi>u</mi>
<mi>k</mi>
</msubsup>
<mo>&CenterDot;</mo>
<mi>log</mi>
<mi> </mi>
<msubsup>
<mi>p</mi>
<mi>u</mi>
<mi>k</mi>
</msubsup>
</mrow>
Wherein, U' represent by the sample from remaining sample set deletion after sample set;Meanwhile, when sample x adds markup information simultaneously
Add after training set, the disaggregated model obtained by study is to the prediction probability of the sample
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710511119.9A CN107247972A (en) | 2017-06-29 | 2017-06-29 | One kind is based on mass-rent technology classification model training method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710511119.9A CN107247972A (en) | 2017-06-29 | 2017-06-29 | One kind is based on mass-rent technology classification model training method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107247972A true CN107247972A (en) | 2017-10-13 |
Family
ID=60014859
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710511119.9A Pending CN107247972A (en) | 2017-06-29 | 2017-06-29 | One kind is based on mass-rent technology classification model training method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107247972A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107808661A (en) * | 2017-10-23 | 2018-03-16 | 中央民族大学 | A kind of Tibetan voice corpus labeling method and system based on collaborative batch Active Learning |
CN108764372A (en) * | 2018-06-08 | 2018-11-06 | Oppo广东移动通信有限公司 | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set |
CN108875821A (en) * | 2018-06-08 | 2018-11-23 | Oppo广东移动通信有限公司 | The training method and device of disaggregated model, mobile terminal, readable storage medium storing program for executing |
CN108921200A (en) * | 2018-06-11 | 2018-11-30 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and medium for classifying to Driving Scene data |
CN109241513A (en) * | 2018-08-27 | 2019-01-18 | 上海宝尊电子商务有限公司 | A kind of method and device based on big data crowdsourcing model data mark |
CN110210294A (en) * | 2019-04-23 | 2019-09-06 | 平安科技(深圳)有限公司 | Evaluation method, device, storage medium and the computer equipment of Optimized model |
CN110321426A (en) * | 2019-07-02 | 2019-10-11 | 腾讯科技(深圳)有限公司 | Abstract abstracting method, device and computer equipment |
WO2020007287A1 (en) * | 2018-07-05 | 2020-01-09 | 第四范式(北京)技术有限公司 | Machine learning process implementation method and apparatus, device, and storage medium |
CN110929807A (en) * | 2019-12-06 | 2020-03-27 | 腾讯科技(深圳)有限公司 | Training method of image classification model, and image classification method and device |
CN111126121A (en) * | 2018-11-01 | 2020-05-08 | 百度在线网络技术(北京)有限公司 | Method, device and equipment for adjusting face recognition model and storage medium |
CN112632179A (en) * | 2019-09-24 | 2021-04-09 | 北京国双科技有限公司 | Model construction method and device, storage medium and equipment |
CN114662659A (en) * | 2022-03-11 | 2022-06-24 | 南京信息工程大学 | Multi-stage transfer learning strategy synthesis-based crowdsourcing text integration method |
-
2017
- 2017-06-29 CN CN201710511119.9A patent/CN107247972A/en active Pending
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107808661B (en) * | 2017-10-23 | 2020-12-11 | 中央民族大学 | Tibetan language voice corpus labeling method and system based on collaborative batch active learning |
CN107808661A (en) * | 2017-10-23 | 2018-03-16 | 中央民族大学 | A kind of Tibetan voice corpus labeling method and system based on collaborative batch Active Learning |
CN108764372A (en) * | 2018-06-08 | 2018-11-06 | Oppo广东移动通信有限公司 | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set |
CN108875821A (en) * | 2018-06-08 | 2018-11-23 | Oppo广东移动通信有限公司 | The training method and device of disaggregated model, mobile terminal, readable storage medium storing program for executing |
CN108764372B (en) * | 2018-06-08 | 2019-07-16 | Oppo广东移动通信有限公司 | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set |
US11138478B2 (en) | 2018-06-08 | 2021-10-05 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for training, classification model, mobile terminal, and readable storage medium |
CN108921200A (en) * | 2018-06-11 | 2018-11-30 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and medium for classifying to Driving Scene data |
US11783590B2 (en) | 2018-06-11 | 2023-10-10 | Apollo Intelligent Driving Technology (Beijing) Co., Ltd. | Method, apparatus, device and medium for classifying driving scenario data |
CN113642633A (en) * | 2018-06-11 | 2021-11-12 | 百度在线网络技术(北京)有限公司 | Method, apparatus, device and medium for classifying driving scene data |
WO2020007287A1 (en) * | 2018-07-05 | 2020-01-09 | 第四范式(北京)技术有限公司 | Machine learning process implementation method and apparatus, device, and storage medium |
CN109241513A (en) * | 2018-08-27 | 2019-01-18 | 上海宝尊电子商务有限公司 | A kind of method and device based on big data crowdsourcing model data mark |
CN111126121A (en) * | 2018-11-01 | 2020-05-08 | 百度在线网络技术(北京)有限公司 | Method, device and equipment for adjusting face recognition model and storage medium |
CN111126121B (en) * | 2018-11-01 | 2023-04-04 | 百度在线网络技术(北京)有限公司 | Method, device and equipment for adjusting face recognition model and storage medium |
CN110210294A (en) * | 2019-04-23 | 2019-09-06 | 平安科技(深圳)有限公司 | Evaluation method, device, storage medium and the computer equipment of Optimized model |
CN110321426A (en) * | 2019-07-02 | 2019-10-11 | 腾讯科技(深圳)有限公司 | Abstract abstracting method, device and computer equipment |
CN110321426B (en) * | 2019-07-02 | 2023-10-27 | 腾讯科技(深圳)有限公司 | Digest extraction method and device and computer equipment |
CN112632179A (en) * | 2019-09-24 | 2021-04-09 | 北京国双科技有限公司 | Model construction method and device, storage medium and equipment |
CN110929807A (en) * | 2019-12-06 | 2020-03-27 | 腾讯科技(深圳)有限公司 | Training method of image classification model, and image classification method and device |
CN114662659A (en) * | 2022-03-11 | 2022-06-24 | 南京信息工程大学 | Multi-stage transfer learning strategy synthesis-based crowdsourcing text integration method |
CN114662659B (en) * | 2022-03-11 | 2022-09-16 | 南京信息工程大学 | Multi-stage transfer learning strategy synthesis-based crowdsourcing text integration method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107247972A (en) | One kind is based on mass-rent technology classification model training method | |
CN104239489B (en) | Utilize the method for similarity searching and improved BP forecast level | |
CN109741332A (en) | A kind of image segmentation and mask method of man-machine coordination | |
CN106528656A (en) | Student history and real-time learning state parameter-based course recommendation realization method and system | |
CN107943784A (en) | Relation extraction method based on generation confrontation network | |
CN106502909B (en) | A kind of aacode defect prediction technique in smart mobile phone application exploitation | |
CN107993140A (en) | A kind of personal credit's methods of risk assessment and system | |
CN105046366B (en) | model training method and device | |
CN103116893B (en) | Digital image labeling method based on multi-exampling multi-marking learning | |
CN102253976B (en) | Metadata processing method and system for spoken language learning | |
KR102203253B1 (en) | Rating augmentation and item recommendation method and system based on generative adversarial networks | |
CN108090520A (en) | Training method, system, device and the readable storage medium storing program for executing of intention assessment model | |
CN104102696A (en) | Content recommendation method and device | |
CN110909125B (en) | Detection method of media rumor of news-level society | |
CN106095973A (en) | The tourism route of a kind of combination short term traffic forecasting recommends method | |
CN106095812A (en) | Intelligent test paper generation method based on similarity measurement | |
CN105841847B (en) | A kind of method for estimating Surface latent heat fluxes | |
CN109165309A (en) | Negative training sample acquisition method, device and model training method, device | |
CN109272003A (en) | A kind of method and apparatus for eliminating unknown error in deep learning model | |
CN109271630A (en) | A kind of intelligent dimension method and device based on natural language processing | |
CN112115993A (en) | Zero sample and small sample evidence photo anomaly detection method based on meta-learning | |
CN107368521A (en) | A kind of Promote knowledge method and system based on big data and deep learning | |
CN105931271A (en) | Behavior locus identification method based on variation BP-HMM | |
CN107943750A (en) | A kind of decomposition convolution method based on WGAN models | |
CN110399547A (en) | For updating the method, apparatus, equipment and storage medium of model parameter |
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: 20171013 |
|
RJ01 | Rejection of invention patent application after publication |