CN108875963A - Optimization method, device, terminal device and the storage medium of machine learning model - Google Patents
Optimization method, device, terminal device and the storage medium of machine learning model Download PDFInfo
- Publication number
- CN108875963A CN108875963A CN201810687251.XA CN201810687251A CN108875963A CN 108875963 A CN108875963 A CN 108875963A CN 201810687251 A CN201810687251 A CN 201810687251A CN 108875963 A CN108875963 A CN 108875963A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- machine learning
- learning model
- trained
- 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
Abstract
The present invention relates to a kind of optimization method of machine learning model, device, terminal device and storage mediums.This method includes：Obtain multiple pending datas；Pending data input machine learning model is filtered out into the pending data for meeting preset condition as labeled data using the machine learning model；Wherein, the labeled data includes training data and verify data；The machine learning model is trained using the training data, determines trained model；According at least to the trained model and the verify data, the machine learning model is updated.The invention enables model can automatic Iterative convergence, man power and material during model optimization is greatly saved.
Description
Technical field
The present invention relates to data processing fields, more particularly to the optimization method, device, terminal of a kind of machine learning model
Equipment and storage medium.
Background technique
With the fast development of artificial intelligence, machine learning is popularized in we live, and is that computer utilizes one
A little training datas obtain machine learning model by training, and learn more useful data information, and use one kind of this model
Method.It has been widely used in our daily lifes, such as：Image recognition, data mining, computer vision, natural language
The different fields such as speech processing, living things feature recognition, voice and handwriting recognition.
Traditional technology needs manually to mark whole data and carries out data in the optimization process to machine learning model
Pretreatment and screening, by adjusting model structure and parameter, and could online use after manual confirmation modelling effect.
Traditional technology expends a large amount of manpowers during machine learning, so that high labor cost, machine learning efficiency
It is low.
Summary of the invention
Based on this, it is necessary to aiming at the problem that high labor cost, machine learning low efficiency in traditional technology, provide a kind of machine
Optimization method, device, terminal device and the storage medium of device learning model.
In a first aspect, the embodiment of the present invention provides a kind of optimization method of machine learning model, including：
Obtain multiple pending datas；
By the pending data input machine learning model, using the machine learning model, filter out meet it is default
The pending data of condition is as labeled data；Wherein, the labeled data includes training data and verify data；
The machine learning model is trained using the training data, determines trained model；
According at least to the trained model and the verify data, the machine learning model is updated.
The pending data is inputted into machine learning model in one of the embodiments, utilizes the machine learning
Model filters out the pending data for meeting preset condition as labeled data, including：
The pending data is inputted into the machine learning model, determines the confidence level of each pending data；
Select confidence level lower than the pending data of preset threshold as labeled data.
The pending data is inputted into the machine learning model in one of the embodiments, is determined each described
The confidence level of pending data, including：
The pending data is inputted into the machine learning model, obtains the preset function of the machine learning model；
According to the preset function, the confidence level of the pending data is determined.
The machine learning model is trained using the training data in one of the embodiments, determines warp
Training pattern, including:
The training data is inputted into the machine learning model, determines corresponding data point in the machine learning model
Cloth variation；
Changed according to the data distribution, determines trained model.
The data distribution changes in one of the embodiments, and the parameter in adjusting training data confirms housebroken
Model, including：
According to the variation of the data distribution, and default optimization algorithm corresponding with the variation of the data distribution, it adjusts
Parameter in the whole training data；
According to the parameter in the training data, trained model is confirmed.
In one of the embodiments, according to the trained model and the verify data, the machine learning is updated
Model, including：
The verify data is inputted into the trained model, determines the first accuracy rate；Wherein, first accuracy rate is
The accuracy rate of the trained model；
According to the machine learning model, the second accuracy rate is determined；Wherein, second accuracy rate is the machine learning
The accuracy rate of model；
Judge whether first accuracy rate is greater than the second accuracy rate：If so, updating the machine learning model for institute
State trained model.
The verify data is inputted into the trained model in one of the embodiments, determines the first accuracy rate, is wrapped
It includes：
The verify data is inputted into the trained model, determines the first model result；
According to artificial labeled data, the second model result is determined；
Second model result and first model result are compared, determine the correct quantity of model prediction and
The quantity that model is predicted in total；
According to the quantity that the correct quantity of the model prediction and the model are predicted in total, the first accuracy rate is determined.
The optimization method of machine learning model provided in this embodiment, terminal device, which passes through, obtains multiple pending datas,
And above-mentioned pending data is inputted into machine learning model, select to meet the pending data of preset condition as labeled data,
Wherein, which includes training data and verify data, is then carried out using above-mentioned training data to machine learning model
Training, determines trained model, and according to trained model, verify data and machine learning model, to update machine learning
Model.It, can be certainly by the way that multiple pending datas are inputted machine learning model since terminal device is when determining labeled data
The low pending data of dynamic screening confidence level, improves the efficiency of determining labeled data, so that mark manpower is largely saved, and
It is trained by the way that the training data in labeled data is inputted machine learning model, enables terminal device to data distribution
Variation timely respond to, be accurately obtained trained model, model automatic Iterative restrained, it is effective to improve training effect
Rate.In addition, can accurately assess the trained model by the way that verify data is inputted trained model, work as determination
When trained model is better than machine learning model, then machine learning model is automatically updated, to be greatly saved in model optimization
Man power and material in the process greatly enhances the efficiency of the update iterative process of machine learning model.
Second aspect, the embodiment of the present invention provide a kind of optimization device of machine learning model, including：
Module is obtained, for obtaining multiple pending datas；
First determining module utilizes the machine learning mould for the pending data to be inputted machine learning model
Type filters out the pending data for meeting preset condition as labeled data；Wherein, the labeled data include training data and
Verify data；
Second determining module is determined for being trained using the training data to the machine learning model through instructing
Experienced model；
Update module, for updating the machine learning model according to the trained model and the verify data.
The third aspect, a kind of terminal device provided in an embodiment of the present invention, including memory and processor, memory storage
There is computer program, the processor realizes following steps when executing computer program：
Obtain multiple pending datas；
By the pending data input machine learning model, using the machine learning model, filter out meet it is default
The pending data of condition is as labeled data；Wherein, the labeled data includes training data and verify data；
The machine learning model is trained using the training data, determines trained model；
According at least to the trained model and the verify data, the machine learning model is updated.
A kind of fourth aspect, computer readable storage medium provided in an embodiment of the present invention, is stored thereon with computer journey
Sequence, the computer program realize following steps when being executed by processor：
Obtain multiple pending datas；
By the pending data input machine learning model, using the machine learning model, filter out meet it is default
The pending data of condition is as labeled data；Wherein, the labeled data includes training data and verify data；
The machine learning model is trained using the training data, determines trained model；
According at least to the trained model and the verify data, the machine learning model is updated.
Optimization method, device, terminal device and the storage medium of machine learning model provided in this embodiment, can pass through
Obtain multiple pending datas, and above-mentioned pending data inputted into machine learning model, selection meet preset condition wait locate
Data are managed as labeled data, wherein the labeled data includes training data and verify data, then utilizes above-mentioned training data
Machine learning model is trained, determines trained model, and according to trained model, verify data and machine learning mould
Type, to update machine learning model.Since terminal device is when determining labeled data, by the way that multiple pending datas are inputted
Machine learning model, can the low pending data of automatic screening confidence level, the efficiency of determining labeled data is improved, thus greatly
Amount saves mark manpower, and is trained by the way that the training data in labeled data is inputted machine learning model, so that eventually
End equipment can timely respond to the variation of data distribution, be accurately obtained trained model, enable model automatic Iterative
Convergence, effectively improves training effectiveness.In addition, by the way that verify data is inputted trained model, it can be accurately to this through instructing
Practice model to be assessed, when determining trained model better than machine learning model, then automatically updates machine learning model, thus
Man power and material during model optimization is greatly saved, greatly enhances the update iteration of machine learning model
The efficiency of process.
Detailed description of the invention
Fig. 1 is a kind of schematic diagram of internal structure for terminal device that one embodiment provides；
Fig. 2 is the flow diagram of the optimization method for the machine learning model that one embodiment provides；
Fig. 3 is the flow diagram of the optimization method for the machine learning model that another embodiment provides；
Fig. 4 is the flow diagram of the optimization method for the machine learning model that another embodiment provides；
Fig. 5 is the flow diagram of the optimization method for the machine learning model that another embodiment provides；
Fig. 6 is the flow diagram of the optimization method for the machine learning model that another embodiment provides；
Fig. 7 is the flow diagram of the optimization method for the machine learning model that another embodiment provides；
Fig. 8 is the flow diagram of the optimization method for the machine learning model that another embodiment provides；
Fig. 9 is the structural schematic diagram of the optimization device for the machine learning model that one embodiment provides；
Figure 10 is the structural schematic diagram of the optimization device for the machine learning model that another embodiment provides；
Figure 11 is the structural schematic diagram of the optimization device for the machine learning model that another embodiment provides；
Figure 12 is the structural schematic diagram of the optimization device for the machine learning model that another embodiment provides.
Specific embodiment
The optimization method of machine learning model provided by the embodiments of the present application can be adapted for terminal device shown in FIG. 1.
The terminal device includes processor, the built-in storage, non-volatile memory medium connected by system bus, this is non-volatile to deposit
Storage media is stored with operating system and computer program, which is operating system and meter in non-volatile memory medium
The step of operation of calculation machine program provides environment, and following methods embodiment can be executed when processor executes the computer program.
Optionally, which can also include network interface, display screen and input unit.Wherein, the processor of the terminal device
For providing calculating and control ability.The network interface of the terminal device is used to communicate with external terminal by network connection.
Optionally, terminal device can be personal computer (personal computer, abbreviation PC), mobile terminal, portable device
Deng the electronic equipment that has data processing function and can interact with external equipment or user, the present embodiment is to terminal device
Concrete form and without limitation.
In traditional technology, in the optimization process to machine learning model, need manually to mark ability after whole data
Data prediction and screening are carried out, and in the training process, by manually adjusting the structure and parameter in model, finally needs people
It online can be used after work confirmation modelling effect.But due to needing to expend a large amount of man power and material in traditional technology, cause
High labor cost, and machine learning low efficiency.The optimization method of machine learning model provided by the embodiments of the present application, device, end
End equipment and readable storage medium storing program for executing aim to solve the problem that the technical problem as above of traditional technology.
In order to which the objects, technical solutions and advantages of the application are more clearly understood, pass through following embodiments and combine attached
Figure, technical solutions in the embodiments of the present application are described in further details.It should be appreciated that specific embodiment described herein
Only to explain the application, it is not used to limit the application.
It should be noted that the executing subject of following methods embodiment can be the optimization device of machine learning model, it should
Device can be implemented as some or all of of terminal device by way of software, hardware or software and hardware combining.It is following
The executing subject of embodiment of the method is illustrated by taking terminal device as an example.
Fig. 2 is the flow diagram of the optimization method for the machine learning model that one embodiment provides, and the present embodiment is related to
Be how terminal device is trained the labeled data in multiple pending datas, determine trained model, and according to warp
Training pattern and verify data update the detailed process of machine learning model.As shown in Fig. 2, this method may include：
S101, multiple pending datas are obtained.
Specifically, pending data is to need to be labeled the data of processing.Multiple pending datas can be including each
The data of seed type, such as：Text data, voice data, image data etc..Above-mentioned pending data can be to be filled by obtaining
The data acquired in real time are set, can be the data downloaded by cloud, can be the various data etc. imported by other equipment,
The present embodiment does not limit this.
Optionally, terminal device, which obtains the process of pending data, to be：Terminal device receives the processing of user's input
Instruction, and pending data is obtained according to process instruction.It wherein, include Data Identification in process instruction.
S102, pending data input machine learning model is filtered out and is met using the machine learning model
The pending data of preset condition is as labeled data；Wherein, the labeled data includes training data and verify data.
Specifically, above-mentioned machine learning model can be model-naive Bayesian, decision-tree model can be, can be and patrol
Regression model is collected, terminal device is input to machine learning mould after getting multiple pending datas, by above-mentioned pending data
In type, machine learning model can choose the pending data for meeting preset condition by handling above-mentioned pending data
As labeled data.Optionally, above-mentioned preset condition, which can be, compares the matching degree of feature in the pending data, can also be with
It is the comparison to parameter in the pending data.By the way that pending data is inputted machine learning model, and model can be certainly
It is dynamic to filter out the low pending data of confidence level, the labeled data for meeting preset condition is selected, mark can be greatly saved
Manpower is labeled processing by the data low to confidence level, can significantly improve the training effect of model.
It should be noted that after determining labeled data, then by the way that manually the labeled data is marked, for example,
Machine learning model is used for the animal identification in image scene, then training data can be multiple images, then handmarking should
Some characteristic informations of animal, wherein this feature information can be the features such as the nose of the animal, eyes.
S103, the machine learning model is trained using the training data, determines trained model.
Specifically, terminal device is after getting labeled data, and the labeled data that will acquire is with 7:3 ratio with
Machine is divided into training data and verify data, wherein training data is for being trained machine learning model, to determine through instructing
Practice model, accuracy rate of the verify data for detection model building, for assessment models.
Machine learning model can be applied in various scenes, such as in classification scene, problem solving scene.Such as：Machine
Device learning model is used to carry out recognition of face to the prospect in image, then training data may include multiple images, these figures
Whether there is face as being marked, or even the position of face is marked.
Optionally, it when being trained with training data to machine learning model, can be calculated by gradient descent method, newton
Method, conjugate gradient method, Cauchy-Newton method scheduling algorithm are trained machine learning model, and then determine trained model.
S104, according at least to the trained model and the verify data, update the machine learning model.
Specifically, trained model is that the model obtained after training passes through verify data after determining trained model
Trained model is verified, and combines the related data of above-mentioned machine learning model, when determining trained modelling effect
When than machine learning model effect optimization, then machine learning model is updated.Optionally, when determining that trained model do not compare machine
When device learning model effect is more excellent, then continues to be trained the trained model, obtain new model, then with model before
Effect is compared, until the new modelling effect be better than before modelling effect when, model before updating is new model.
The optimization method of machine learning model provided in an embodiment of the present invention, terminal device is by obtaining multiple numbers to be processed
According to, and above-mentioned pending data is inputted into machine learning model, select the pending data for meeting preset condition as mark number
Include training data and verify data according to, wherein the labeled data, then using above-mentioned training data to machine learning model into
Row training, determines trained model, and according to trained model, verify data and machine learning model, to update engineering
Practise model.It, can by the way that multiple pending datas are inputted machine learning model since terminal device is when determining labeled data
The low pending data of automatic screening confidence level, improves the efficiency of determining labeled data, so that mark manpower is largely saved,
And be trained by the way that the training data in labeled data is inputted machine learning model, enable terminal device to data point
The variation of cloth timely responds to, and is accurately obtained trained model, and model automatic Iterative is restrained, effective to improve training
Efficiency.In addition, can accurately be assessed the trained model, by the way that verify data is inputted trained model when true
When fixed trained model is better than machine learning model, then machine learning model is automatically updated, to be greatly saved excellent in model
Man power and material during change greatly enhances the efficiency of the update iterative process of machine learning model.
Fig. 3 is the optimization method flow diagram for the machine learning model that another embodiment provides.What the present embodiment was related to
It is how pending data is inputted machine learning model by terminal device, selects the detailed process of labeled data.Based on above-mentioned Fig. 2
On the basis of illustrated embodiment, as shown in figure 3, a kind of achievable mode of S101 may include:
S201, the pending data is inputted into the machine learning model, determines setting for each pending data
Reliability.
Specifically, the confidence level of pending data is the probability that true value occurs in confidence interval.When terminal device obtains
When to multiple pending datas, and the pending data is inputted into above-mentioned machine learning model, determines above-mentioned pending data
Confidence level optionally when determining the confidence level of each pending data, can determine each number to be processed by the law of large numbers
According to confidence level.
Optionally, as shown in figure 4, a kind of possible implementation of S201 includes：
S301, the pending data is inputted into the machine learning model, obtains the default of the machine learning model
Function.
Specifically, the pending data is inputted above-mentioned machine learning model, can by calling model database come
The preset function of machine learning model is got, it can also be by the way that according to above-mentioned machine learning model, corresponding acquisition be therein
Preset function.Wherein, the corresponding preset function of different machine learning models is different.
S302, according to the preset function, determine the confidence level of the pending data.
Specifically, preset function is the function for determining confidence level in machine learning model, when multiple pending datas are defeated
After entering machine learning model, preset function therein is got, which is substituted into the preset function, can be determined
The confidence level of each pending data.Wherein, the corresponding confidence level of different pending datas may be different, it is also possible to identical.
For example, then pending data can be more when machine learning model is used for animal identification in image scene
A image, the confidence level of the pending data are the probability that each image is the animal.
The optimization method of machine learning model provided by the above embodiment, by the way that above-mentioned pending data is inputted engineering
Model is practised, after the preset function for obtaining above-mentioned machine learning model, and according to the preset function, determines the confidence of pending data
Degree.By obtaining preset function, so as to accurately obtain the confidence level of pending data, mark is further accurately determined
Data.
S202, select confidence level lower than the pending data of preset threshold as labeled data.
Specifically, being carried out after the confidence level for determining each pending data, and by above-mentioned confidence level and preset threshold
Comparison, optionally, each confidence level can be compared with preset threshold respectively, first can also mutually be compared confidence level
Compared with determining the size relation of each confidence level, wherein can arrange above-mentioned confidence level by way of from big to small
Sequence can also determine the relationship between confidence level by bubble sort to above-mentioned confidence level, and by the confidence level and default threshold
Value is compared, so that the pending data that confidence level is lower than preset threshold is chosen, as labeled data.
The optimization method of machine learning model provided in this embodiment, terminal device is by inputting machine for pending data
Learning model gets the preset function of above-mentioned machine learning model, and according to preset function, determines each pending data
Then the confidence level of each pending data is compared by confidence level with preset threshold, determine confidence level lower than preset threshold
Pending data as labeled data.Since the terminal device is when selecting labeled data, using confidence level as criterion,
And by the way that confidence level to be compared with preset threshold, to accurately determine labeled data, model is further improved
Training effectiveness.
Fig. 5 is the optimization method flow diagram for the machine learning model that another embodiment provides.What the present embodiment was related to
It is how terminal device is trained machine learning model using training data, determines the detailed process of trained model.Base
On the basis of above-mentioned embodiment illustrated in fig. 2, as shown in figure 5, a kind of achievable mode of S103 may include:
S401, the training data is inputted into the machine learning model, determined corresponding in the machine learning model
Data distribution variation.
Specifically, terminal device is by inputting machine learning model for pending data, after determining labeled data, and should
Labeled data is randomly divided into two groups of data：Training data and verify data, it is after determining training data, above-mentioned training data is defeated
Enter in machine learning model, optionally, the machine learning model can by continuous study, by in the machine learning model
Some ATTRIBUTE INDEXs be compared, determine corresponding data distribution variation in machine learning model.
Optionally, can by measure the data central tendency index come determine the data distribution in training data become
Change, wherein the index for measuring the central tendency of the data can be location average, can be digital average number, can also lead to
Data profile is crossed to determine that the data distribution in training data changes.
S402, changed according to the data distribution, the parameter in adjusting training data determines trained model.
Specifically, data distribution variation is the central tendency of data distribution, terminal device is in study into training data
It, can be according to the variation of the data distribution, by relationship, trend and the mode in study identification data, to adjust when changes in distribution
Parameter in training data determines trained model.
Optionally, as shown in fig. 6, a kind of possible implementation of S402 includes：
S501, according to the variation of the data distribution, and default optimization corresponding with the variation of the data distribution is calculated
Method adjusts the parameter in the training data.
Specifically, machine learning model is in the training process, it can be for the training data for having data variation, by some
Optimization algorithm adjusts parameter therein.
It should be noted that being directed to different machine learning models, the variation of data distribution is different, it is corresponding preset it is excellent
It is also different to change algorithm.Optionally, machine learning is divided into supervised learning, unsupervised learning and semi-supervised learning.Wherein, for supervision
Problem concerning study, such as classification problem can solve classification problem using sorting algorithm, by dividing known class training set
Analysis, is therefrom found classifying rules, the classification of new data is predicted with this；Wherein, when machine learning model is model-naive Bayesian
When, it can be by parameters such as weights in NB Algorithm adjusting training data, when machine learning model is supporting vector
When model, the supporting vector in support vector machines, such as regression problem, then machine learning can be adjusted by algorithm of support vector machine
, can be by the parameter in linear regression algorithm adjusting training data when model is Logic Regression Models, which, which can be, patrols
Collect the coefficient in returning.
S502, according to the parameter in the training data, confirm trained model.
Wherein, it after adjusting the parameter in training data, can be determined according to the parameter in the training data through instructing
Practice model.
The optimization method of machine learning model in above-described embodiment, machine learning model in the training process, pass through root
According to training data, terminal device is enabled to learn automatically to corresponding data distribution to change, and then inside adjusting training data
Parameter, eliminate the adjustment manually to model, so that model is adjusted and be optimized automatically, greatly enhance pair
The training effectiveness of model.
The optimization method of machine learning model provided in this embodiment, terminal device is by inputting engineering for training data
Model is practised, wherein corresponding data distribution variation is determined, and change according to the data distribution, passes through optimization algorithm, adjusting training
Parameter in data, so that it is determined that trained model.Since the terminal device is can be automatic right according to the variation of data distribution
The parameter in model is adjusted, to optimize to model, further accurately determination provides trained model, is greatly saved
Manpower significantly improves the training effect to model.
Fig. 7 is the optimization method flow diagram for the machine learning model that another embodiment provides.What the present embodiment was related to
How it is terminal device according to trained model, verify data and machine learning model, updates the specific mistake of machine learning model
Journey.On the basis of above-mentioned embodiment illustrated in fig. 2, as shown in fig. 7, a kind of achievable mode of S104 may include:
S601, the verify data is inputted into the trained model, determines the first accuracy rate；Wherein, described first is quasi-
True rate is the accuracy rate of the trained model.
Specifically, verify data be used for testing model, terminal device after determining labeled data, and by labeled data with
Machine is divided into two groups of data, and one group is training data, and one group is verify data, wherein by training data to machine learning model
After the completion of training, trained model is obtained, and verify data is inputted into the trained model, to determine the accuracy rate of the model.
It optionally, can be by sample number and total number of samples that classifier correctly separates for given test data set
The ratio between, determine the accuracy rate of trained model.
Optionally, as shown in figure 8, a kind of possible implementation of above-mentioned S601 may include:
S701, the verify data is inputted into the trained model, determines the first model result.
Specifically, can determine whether out corresponding first model result after verify data is inputted trained model.
For example, when the machine learning model is used for the identification to animal, then verify data is multiple images, when verifying number
After inputting trained model, terminal device can be automatically by the relevant treatment of the model, to conclude and identify specific mesh
Mark, i.e., conclude multiple images and learnt, to identify the animal in the image, as the first model knot of the model
Fruit.
S702, according to artificial labeled data, determine the second model result.
For example, machine learning model is for when identifying to animal, then terminal device to select labeled data, and passes through people
Work is labeled the data, wherein the result of the labeled data is manually to be obtained according to autognosis to the image recognition
As a result, as the second model result.
S703, second model result and first model result are compared, determines that model prediction is correct
The quantity that quantity and model are predicted in total.
For example, the first model result is engineering when machine learning model is used for animal identification in image scene
The trained rear new image comprising target animal feature formed of model is practised, the second model result is manually to mark comprising mesh
The original image of animal character is marked, the quantity that the model in the scene is predicted in total is the key point of all features of the target animal
Quantity, the correct quantity of model prediction are the first model result and the second model knot about in all key points of target person
The identical keypoint quantity of feature in fruit.Optionally, vector or rectangular can be passed through to the key point of target animal in image
Formula carries out certain digitized processing and obtains the first model result and the second model result.
S704, the quantity predicted in total according to the correct quantity of model prediction and the model, determine the first accuracy rate.
Further, after determining the quantity that the correct quantity of model prediction and model are predicted in total, by the model prediction
The ratio for the quantity that correct quantity and model are predicted in total, is determined as the first accuracy rate.First accuracy rate is as above-mentioned warp
The judgment criteria of the accuracy rate of training pattern can preferably assess the effect of the trained model.
The optimization method of machine learning model provided by the above embodiment, by the way that verify data is inputted trained model,
The first model result is obtained, and is compared with the second obtained model result is manually marked, can accurately determine the warp
The accuracy rate of training pattern.
S602, according to the machine learning model, determine the second accuracy rate；Wherein, second accuracy rate is the machine
The accuracy rate of device learning model；
Specifically, the accuracy rate by determining the model, assesses above-mentioned machine learning model, with above-mentioned determination the
The process of one accuracy rate is identical, and verify data is inputted the machine learning model, can obtain corresponding model result, and with it is artificial
The result of mark is compared, so that it is determined that the second accuracy rate.Wherein, second accuracy rate is as judgement machine learning model effect
The condition of fruit.
S603, judge whether first accuracy rate is greater than the second accuracy rate：If so, updating the machine learning model
For the trained model.
Specifically, after terminal device obtains the first accuracy rate and the second accuracy rate, the first accuracy rate is accurate with second
Rate is compared, and when the first accuracy rate is greater than the second accuracy rate, then updating machine learning model is trained model；When first
When accuracy rate is less than or equal to the second accuracy rate, then the trained model is trained again, and then by learning it automatically
In data distribution variation, new model is obtained, thus continuous iteration optimization, until the accuracy rate ratio of obtained new model
When the accuracy rate of model is big before, then model is new model before automatically updating.It is judged automatically by terminal device trained
The modelling effect of model realizes the automatic Iterative to machine-learning process, substantially increases the effect of model modification iterative process
Rate.
The optimization method of machine learning model provided in this embodiment is obtained by the way that verify data is inputted trained model
Machine learning model is inputted to the first accuracy rate, and using same verify data, the second accuracy rate is obtained, passes through comparison first
When accuracy rate is greater than the second accuracy rate, then updating machine learning model is trained model.Terminal device is trained by determination
The accuracy rate of model and machine learning model, using accuracy rate as the standard of decision model effect, to improve testing for model
Demonstrate,prove efficiency；Meanwhile when determining the first accuracy rate better than the second accuracy rate, terminal device can automatically to machine learning model into
Row optimization, avoids the influence manually generated to model optimization.
Fig. 9 is the structural schematic diagram of the optimization device for the machine learning model that an embodiment provides.As shown in figure 9, the machine
The optimization device of device learning model may include obtaining module 11, the first determining module 12, the second determining module 13 and updating mould
Block 14.
Specifically, module 11 is obtained, for obtaining multiple pending datas.
First determining module 12 utilizes the machine learning for the pending data to be inputted machine learning model
Model filters out the pending data for meeting preset condition as labeled data；Wherein, the labeled data includes training data
And verify data；
Second determining module 13 determines warp for being trained using the training data to the machine learning model
Trained model；
Update module 14, for updating the machine learning according at least to the trained model and the verify data
Model.
The optimization device of machine learning model provided in this embodiment can execute above method embodiment, realize former
Reason is similar with technical effect, and details are not described herein.
Figure 10 is the structural schematic diagram of the optimization device for the machine learning model that another embodiment provides.In above-mentioned Fig. 9 institute
On the basis of showing embodiment, optionally, as shown in Figure 10, above-mentioned first determining module 12 may include the first determination unit 121,
Comparing unit 122 and selecting unit 123.
First determination unit 121 determines each described for the pending data to be inputted the machine learning model
The confidence level of pending data.
Comparing unit 122, for by the confidence level of the pending data compared with preset threshold.
Selecting unit 123, for selecting confidence level lower than the pending data of preset threshold as labeled data.
Above-mentioned first determination unit 121 in one of the embodiments, is specifically used for the pending data inputting institute
Machine learning model is stated, the preset function of the machine learning model is obtained；According to the preset function, determine described to be processed
The confidence level of data.
The optimization device of machine learning model provided in this embodiment can execute above method embodiment, realize former
Reason is similar with technical effect, and details are not described herein.
Figure 11 is the structural schematic diagram of the optimization device for the machine learning model that another embodiment provides.In above-mentioned Figure 10 institute
On the basis of showing embodiment, optionally, as shown in figure 11, above-mentioned second determining module 13 may include the second determination unit 131
With third determination unit 132.
Second determination unit 131 determines the engineering for the training data to be inputted the machine learning model
Practise corresponding data distribution variation in model.
Third determination unit 132, for being changed according to the data distribution, the parameter in adjusting training data determines warp
Training pattern.
Above-mentioned third determination unit 132 in one of the embodiments, specifically for the change according to the data distribution
Change, and default optimization algorithm corresponding with the variation of the data distribution, adjusts the parameter in the training data；According to institute
The parameter in training data is stated, confirms trained model.
The optimization device of machine learning model provided in this embodiment can execute above method embodiment, realize former
Reason is similar with technical effect, and details are not described herein.
Figure 12 is the structural schematic diagram of the optimization device for the machine learning model that another embodiment provides.In above-mentioned Figure 10 institute
On the basis of showing embodiment, optionally, as shown in figure 12, above-mentioned update module 14 may include the 4th determination unit the 141, the 5th
Determination unit 142 and updating unit 143.
4th determination unit 141 determines the first accuracy rate for the verify data to be inputted the trained model；
Wherein, first accuracy rate is the accuracy rate of the trained model；
5th determination unit 142, for that will determine the second accuracy rate according to the machine learning model；Wherein, described
Two accuracys rate are the accuracy rate of the machine learning model；
Updating unit 143, for judging whether first accuracy rate is greater than the second accuracy rate：If so, described in updating
Machine learning model is the trained model.
Above-mentioned 5th determination unit 142 in one of the embodiments, being specifically used for will be described in verify data input
Trained model determines the first model result；According to artificial labeled data, the second model result is determined；By second model
As a result it is compared with first model result, determines the quantity that the correct quantity of model prediction and model are predicted in total；Root
According to the quantity that the correct quantity of the model prediction and the model are predicted in total, the first accuracy rate is determined.
The specific restriction of optimization device about machine learning model may refer to above for machine learning model
The restriction of optimization method, details are not described herein.Modules in the optimization device of above-mentioned machine learning model can whole or portion
Divide and is realized by software, hardware and combinations thereof.Above-mentioned each module can be embedded in the form of hardware or independently of in terminal device
Processor in, can also be stored in a software form in the memory in terminal device, in order to processor call execute with
The corresponding operation of upper modules.
In one embodiment, a kind of terminal device is provided, which can be terminal, and internal structure chart can
With as shown in Figure 1.The terminal device includes processor, memory, network interface, the display screen and defeated connected by system bus
Enter device.Wherein, the processor of the terminal device is for providing calculating and control ability.The memory of the terminal device includes non-
Volatile storage medium, built-in storage.The non-volatile memory medium is stored with operating system and computer program.The interior storage
Device provides environment for the operation of operating system and computer program in non-volatile memory medium.The network of the terminal device connects
Mouth with external terminal by network connection for being communicated.To realize at a kind of image when the computer program is executed by processor
Reason method.The display screen of the terminal device can be liquid crystal display or electric ink display screen, the input of the terminal device
Device can be the touch layer covered on display screen, be also possible to the key being arranged on terminal device shell, trace ball or touch-control
Plate can also be external keyboard, Trackpad or mouse etc..
It will be understood by those skilled in the art that structure shown in Fig. 1, only part relevant to application scheme is tied
The block diagram of structure, does not constitute the restriction for the terminal device being applied thereon to application scheme, and specific terminal device can be with
Including than more or fewer components as shown in the figure, perhaps combining certain components or with different component layouts.
In one embodiment, a kind of terminal device, including memory and processor are provided, the memory is stored with
Computer program, the processor realize following steps when executing the computer program：
Obtain multiple pending datas；
By the pending data input machine learning model, using the machine learning model, filter out meet it is default
The pending data of condition is as labeled data；Wherein, the labeled data includes training data and verify data；
The machine learning model is trained using the training data, determines trained model；
According at least to the trained model and the verify data, the machine learning model is updated.
In one embodiment, following steps are also realized when processor executes computer program：
The pending data is inputted into the machine learning model, determines the confidence level of each pending data；
Select confidence level lower than the pending data of preset threshold as labeled data.
In one embodiment, following steps are also realized when processor executes computer program：
The pending data is inputted into the machine learning model, obtains the preset function of the machine learning model；
According to the preset function, the confidence level of the pending data is determined.
In one embodiment, following steps are also realized when processor executes computer program：
The training data is inputted into the machine learning model, determines corresponding data point in the machine learning model
Cloth variation；Changed according to the data distribution, the parameter in adjusting training data determines trained model.
In one embodiment, following steps are also realized when processor executes computer program：
According to the variation of the data distribution, and default optimization algorithm corresponding with the variation of the data distribution, it adjusts
Parameter in the whole training data；According to the parameter in the training data, trained model is confirmed.
In one embodiment, following steps are also realized when processor executes computer program：
The verify data is inputted into the trained model, determines the first accuracy rate；According to the machine learning model,
Determine the second accuracy rate；Judge whether first accuracy rate is greater than the second accuracy rate：If so, updating the machine learning mould
Type is the trained model.Wherein, first accuracy rate is the accuracy rate of the trained model, second accuracy rate
For the accuracy rate of the machine learning model；
In one embodiment, following steps are also realized when processor executes computer program：
The verify data is inputted into the trained model, determines the first model result；According to artificial labeled data, really
Fixed second model result；Second model result and first model result are compared, determine that model prediction is correct
Quantity and the quantity predicted in total of model；The number predicted in total according to the correct quantity of the model prediction and the model
Amount, determines the first accuracy rate.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program realizes following steps when being executed by processor：
Obtain multiple pending datas；
By the pending data input machine learning model, using the machine learning model, filter out meet it is default
The pending data of condition is as labeled data；Wherein, the labeled data includes training data and verify data；
The machine learning model is trained using the training data, determines trained model；
According at least to the trained model and the verify data, the machine learning model is updated.
In one embodiment, following steps are also realized when computer program is executed by processor：
The pending data is inputted into the machine learning model, determines the confidence level of each pending data；
Select confidence level lower than the pending data of preset threshold as labeled data；Wherein, the preset condition is the confidence level
Lower than preset threshold.
In one embodiment, following steps are also realized when computer program is executed by processor：
The pending data is inputted into the machine learning model, obtains the preset function of the machine learning model；
According to the preset function, the confidence level of the pending data is determined.
In one embodiment, following steps are also realized when computer program is executed by processor：
The training data is inputted into the machine learning model, determines corresponding data point in the machine learning model
Cloth variation；Changed according to the data distribution, the parameter in adjusting training data determines trained model.
In one embodiment, following steps are also realized when computer program is executed by processor：
According to the variation of the data distribution, and default optimization algorithm corresponding with the variation of the data distribution, it adjusts
Parameter in the whole training data；According to the parameter in the training data, trained model is confirmed.
In one embodiment, following steps are also realized when computer program is executed by processor：
The verify data is inputted into the trained model, determines the first accuracy rate；According to the machine learning model,
Determine the second accuracy rate；Judge whether first accuracy rate is greater than the second accuracy rate：If so, updating the machine learning mould
Type is the trained model.Wherein, first accuracy rate is the accuracy rate of the trained model；Second accuracy rate
For the accuracy rate of the machine learning model.
In one embodiment, following steps are also realized when computer program is executed by processor：
The verify data is inputted into the trained model, determines the first model result；According to artificial labeled data, really
Fixed second model result；Second model result and first model result are compared, determine that model prediction is correct
Quantity and the quantity predicted in total of model；The number predicted in total according to the correct quantity of the model prediction and the model
Amount, determines the first accuracy rate.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection of the invention
Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (10)
1. a kind of optimization method of machine learning model, which is characterized in that including：
Obtain multiple pending datas；
Pending data input machine learning model is filtered out using the machine learning model and meets preset condition
Pending data as labeled data；Wherein, the labeled data includes training data and verify data；
The machine learning model is trained using the training data, determines trained model；
According at least to the trained model and the verify data, the machine learning model is updated.
2. the method according to claim 1, wherein the pending data is inputted machine learning model, benefit
With the machine learning model, the pending data for meeting preset condition is filtered out as labeled data, including：
The pending data is inputted into the machine learning model, determines the confidence level of each pending data；
Select confidence level lower than the pending data of preset threshold as labeled data.
3. according to the method described in claim 2, it is characterized in that, the pending data is inputted the machine learning mould
Type determines the confidence level of each pending data, including：
The pending data is inputted into the machine learning model, obtains the preset function of the machine learning model；
According to the preset function, the confidence level of the pending data is determined.
4. method according to claim 1-3, which is characterized in that using the training data to the engineering
It practises model to be trained, determines trained model, including:
The training data is inputted into the machine learning model, determines that corresponding data distribution becomes in the machine learning model
Change；
Changed according to the data distribution, the parameter in adjusting training data determines trained model.
5. according to the method described in claim 4, it is characterized in that, being changed according to the data distribution, in adjusting training data
Parameter, confirm housebroken model, including：
According to the variation of the data distribution, and default optimization algorithm corresponding with the variation of the data distribution, adjust institute
State the parameter in training data；
According to the parameter in the training data, trained model is confirmed.
6. method according to claim 1-3, which is characterized in that according to the trained model and the verifying
Data update the machine learning model, including：
The verify data is inputted into the trained model, determines the first accuracy rate；Wherein, first accuracy rate is described
The accuracy rate of trained model；
According to the machine learning model, the second accuracy rate is determined；Wherein, second accuracy rate is the machine learning model
Accuracy rate；
Judge whether first accuracy rate is greater than the second accuracy rate：If so, updating the machine learning model is the warp
Training pattern.
7. according to the method described in claim 6, it is characterized in that, the verify data is inputted the trained model, really
Fixed first accuracy rate, including：
The verify data is inputted into the trained model, determines the first model result；
According to artificial labeled data, the second model result is determined；
Second model result and first model result are compared, determine the correct quantity of model prediction and model
The quantity predicted in total；
According to the quantity that the correct quantity of the model prediction and the model are predicted in total, the first accuracy rate is determined.
8. a kind of optimization device of machine learning model, which is characterized in that including：
Module is obtained, for obtaining multiple pending datas；
First determining module utilizes the machine learning model, sieve for the pending data to be inputted machine learning model
The pending data for meeting preset condition is selected as labeled data；Wherein, the labeled data includes training data and verifying
Data；
Second determining module is determined housebroken for being trained using the training data to the machine learning model
Model；
Update module, for updating the machine learning model according to the trained model and the verify data.
9. a kind of terminal device, including memory and processor, the memory are stored with computer program, which is characterized in that
The processor realizes the step of any one of claim 1-7 the method when executing the computer program.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of any one of claim 1-7 the method is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810687251.XA CN108875963A (en) | 2018-06-28 | 2018-06-28 | Optimization method, device, terminal device and the storage medium of machine learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810687251.XA CN108875963A (en) | 2018-06-28 | 2018-06-28 | Optimization method, device, terminal device and the storage medium of machine learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108875963A true CN108875963A (en) | 2018-11-23 |
Family
ID=64296216
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810687251.XA Pending CN108875963A (en) | 2018-06-28 | 2018-06-28 | Optimization method, device, terminal device and the storage medium of machine learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108875963A (en) |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583590A (en) * | 2018-11-29 | 2019-04-05 | 深圳和而泰数据资源与云技术有限公司 | Data processing method and data processing equipment |
| CN109784343A (en) * | 2019-01-25 | 2019-05-21 | 上海深杳智能科技有限公司 | A kind of resource allocation methods and terminal based on deep learning model |
| CN109800441A (en) * | 2019-02-01 | 2019-05-24 | 北京金山数字娱乐科技有限公司 | A kind of model output recommended method and device, model export recommender system |
| CN109815991A (en) * | 2018-12-29 | 2019-05-28 | 北京城市网邻信息技术有限公司 | Training method, device, electronic equipment and the storage medium of machine learning model |
| CN109829375A (en) * | 2018-12-27 | 2019-05-31 | 深圳云天励飞技术有限公司 | A kind of machine learning method, device, equipment and system |
| CN109934352A (en) * | 2019-03-06 | 2019-06-25 | 北京深度奇点科技有限公司 | The automatic evolvement method of model of mind |
| CN110334633A (en) * | 2019-06-27 | 2019-10-15 | 北京御航智能科技有限公司 | Identification inspection data and the method, apparatus and storage medium for updating identification model |
| CN110334816A (en) * | 2019-07-12 | 2019-10-15 | 深圳市智物联网络有限公司 | A kind of industrial equipment detection method, device, equipment and readable storage medium storing program for executing |
| CN110797101A (en) * | 2019-10-28 | 2020-02-14 | 腾讯医疗健康(深圳)有限公司 | Medical data processing method, device, readable storage medium and computer equipment |
| CN110909887A (en) * | 2019-11-21 | 2020-03-24 | 广州极飞科技有限公司 | Model optimization method and device |
| CN111008706A (en) * | 2019-12-09 | 2020-04-14 | 长春嘉诚信息技术股份有限公司 | Processing method for automatically labeling, training and predicting mass data |
| CN111046935A (en) * | 2019-12-05 | 2020-04-21 | 北京公瑾科技有限公司 | Receipt generation method and device based on machine learning model |
| CN111209931A (en) * | 2019-12-23 | 2020-05-29 | 深圳智链物联科技有限公司 | Data processing method, platform, terminal device and storage medium |
| CN111309706A (en) * | 2020-01-20 | 2020-06-19 | 北京明略软件系统有限公司 | Model training method and device, readable storage medium and electronic equipment |
| CN111340054A (en) * | 2018-12-18 | 2020-06-26 | 北京嘀嘀无限科技发展有限公司 | Data labeling method and device and data processing equipment |
| CN111401563A (en) * | 2018-12-28 | 2020-07-10 | 杭州海康威视数字技术股份有限公司 | Machine learning model updating method and device |
| CN111708810A (en) * | 2020-06-17 | 2020-09-25 | 北京世纪好未来教育科技有限公司 | Model optimization recommendation method and device and computer storage medium |
| CN111814819A (en) * | 2019-04-11 | 2020-10-23 | 黑芝麻智能科技(重庆)有限公司 | Hybrid data labeling method for machine learning |
| CN112070238A (en) * | 2020-11-10 | 2020-12-11 | 鹏城实验室 | Accurate machine learning asynchronous prediction method and system and storage medium |
| CN112598326A (en) * | 2020-12-31 | 2021-04-02 | 五八有限公司 | Model iteration method and device, electronic equipment 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 |
| CN112805725A (en) * | 2020-01-06 | 2021-05-14 | 深圳市微蓝智能科技有限公司 | Data processing method and device and computer readable storage medium |
| CN113536004A (en) * | 2021-06-25 | 2021-10-22 | 猫岐智能科技（上海）有限公司 | Real-time graph monitoring system and method |
| US20210334702A1 (en) * | 2020-04-28 | 2021-10-28 | Mitsubishi Heavy Industries, Ltd. | Model evaluating device, model evaluating method, and program |
| US20210334440A1 (en) * | 2020-04-28 | 2021-10-28 | Mitsubishi Heavy Industries, Ltd. | Model optimization device, model optimization method, and program |
| WO2022213565A1 (en) * | 2021-04-07 | 2022-10-13 | 全球能源互联网研究院有限公司 | Review method and apparatus for prediction result of artificial intelligence model |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105912500A (en) * | 2016-03-30 | 2016-08-31 | 百度在线网络技术（北京）有限公司 | Machine learning model generation method and machine learning model generation device |
| CN106156859A (en) * | 2016-06-17 | 2016-11-23 | 江苏大学 | A kind of used container mechanics output Forecasting Methodology |
| US20160350675A1 (en) * | 2015-06-01 | 2016-12-01 | Facebook, Inc. | Systems and methods to identify objectionable content |
| CN106571135A (en) * | 2016-10-27 | 2017-04-19 | 苏州大学 | Whisper speech feature extraction method and system |
| CN106650815A (en) * | 2016-12-28 | 2017-05-10 | 北京航天测控技术有限公司 | Dynamic particle swarm optimization-based multi-core support vector machine pod fault diagnosis method and apparatus |
| CN107180026A (en) * | 2017-05-02 | 2017-09-19 | 苏州大学 | The event phrase learning method and device of a kind of word-based embedded Semantic mapping |
| CN107392125A (en) * | 2017-07-11 | 2017-11-24 | 中国科学院上海高等研究院 | Training method/system, computer-readable recording medium and the terminal of model of mind |
-
2018
- 2018-06-28 CN CN201810687251.XA patent/CN108875963A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160350675A1 (en) * | 2015-06-01 | 2016-12-01 | Facebook, Inc. | Systems and methods to identify objectionable content |
| CN105912500A (en) * | 2016-03-30 | 2016-08-31 | 百度在线网络技术（北京）有限公司 | Machine learning model generation method and machine learning model generation device |
| CN106156859A (en) * | 2016-06-17 | 2016-11-23 | 江苏大学 | A kind of used container mechanics output Forecasting Methodology |
| CN106571135A (en) * | 2016-10-27 | 2017-04-19 | 苏州大学 | Whisper speech feature extraction method and system |
| CN106650815A (en) * | 2016-12-28 | 2017-05-10 | 北京航天测控技术有限公司 | Dynamic particle swarm optimization-based multi-core support vector machine pod fault diagnosis method and apparatus |
| CN107180026A (en) * | 2017-05-02 | 2017-09-19 | 苏州大学 | The event phrase learning method and device of a kind of word-based embedded Semantic mapping |
| CN107392125A (en) * | 2017-07-11 | 2017-11-24 | 中国科学院上海高等研究院 | Training method/system, computer-readable recording medium and the terminal of model of mind |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583590B (en) * | 2018-11-29 | 2020-11-13 | 深圳和而泰数据资源与云技术有限公司 | Data processing method and data processing device |
| CN109583590A (en) * | 2018-11-29 | 2019-04-05 | 深圳和而泰数据资源与云技术有限公司 | Data processing method and data processing equipment |
| CN111340054A (en) * | 2018-12-18 | 2020-06-26 | 北京嘀嘀无限科技发展有限公司 | Data labeling method and device and data processing equipment |
| CN109829375A (en) * | 2018-12-27 | 2019-05-31 | 深圳云天励飞技术有限公司 | A kind of machine learning method, device, equipment and system |
| CN111401563A (en) * | 2018-12-28 | 2020-07-10 | 杭州海康威视数字技术股份有限公司 | Machine learning model updating method and device |
| CN111401563B (en) * | 2018-12-28 | 2023-11-03 | 杭州海康威视数字技术股份有限公司 | Machine learning model updating method and device |
| CN109815991A (en) * | 2018-12-29 | 2019-05-28 | 北京城市网邻信息技术有限公司 | Training method, device, electronic equipment and the storage medium of machine learning model |
| CN109815991B (en) * | 2018-12-29 | 2021-02-19 | 北京城市网邻信息技术有限公司 | Training method and device of machine learning model, electronic equipment and storage medium |
| CN109784343B (en) * | 2019-01-25 | 2023-05-12 | 上海深杳智能科技有限公司 | Resource allocation method and terminal based on deep learning model |
| CN109784343A (en) * | 2019-01-25 | 2019-05-21 | 上海深杳智能科技有限公司 | A kind of resource allocation methods and terminal based on deep learning model |
| CN109800441A (en) * | 2019-02-01 | 2019-05-24 | 北京金山数字娱乐科技有限公司 | A kind of model output recommended method and device, model export recommender system |
| CN109800441B (en) * | 2019-02-01 | 2023-03-24 | 北京金山数字娱乐科技有限公司 | Model output recommendation method and device and model output recommendation system |
| CN109934352A (en) * | 2019-03-06 | 2019-06-25 | 北京深度奇点科技有限公司 | The automatic evolvement method of model of mind |
| CN111814819B (en) * | 2019-04-11 | 2024-02-20 | 黑芝麻智能科技(重庆)有限公司 | Mixed data marking method for machine learning |
| CN111814819A (en) * | 2019-04-11 | 2020-10-23 | 黑芝麻智能科技(重庆)有限公司 | Hybrid data labeling method for machine learning |
| CN110334633A (en) * | 2019-06-27 | 2019-10-15 | 北京御航智能科技有限公司 | Identification inspection data and the method, apparatus and storage medium for updating identification model |
| CN110334816A (en) * | 2019-07-12 | 2019-10-15 | 深圳市智物联网络有限公司 | A kind of industrial equipment detection method, device, equipment and readable storage medium storing program for executing |
| 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 |
| CN110797101B (en) * | 2019-10-28 | 2023-11-03 | 腾讯医疗健康(深圳)有限公司 | Medical data processing method, medical data processing device, readable storage medium and computer equipment |
| CN110797101A (en) * | 2019-10-28 | 2020-02-14 | 腾讯医疗健康(深圳)有限公司 | Medical data processing method, device, readable storage medium and computer equipment |
| CN110909887A (en) * | 2019-11-21 | 2020-03-24 | 广州极飞科技有限公司 | Model optimization method and device |
| CN110909887B (en) * | 2019-11-21 | 2023-06-09 | 广州极飞科技股份有限公司 | Model optimization method and device |
| CN111046935B (en) * | 2019-12-05 | 2023-08-01 | 北京子敬科技有限公司 | Document generation credential method and device based on machine learning model |
| CN111046935A (en) * | 2019-12-05 | 2020-04-21 | 北京公瑾科技有限公司 | Receipt generation method and device based on machine learning model |
| CN111008706B (en) * | 2019-12-09 | 2023-05-05 | 长春嘉诚信息技术股份有限公司 | Processing method for automatically labeling, training and predicting mass data |
| CN111008706A (en) * | 2019-12-09 | 2020-04-14 | 长春嘉诚信息技术股份有限公司 | Processing method for automatically labeling, training and predicting mass data |
| CN111209931A (en) * | 2019-12-23 | 2020-05-29 | 深圳智链物联科技有限公司 | Data processing method, platform, terminal device and storage medium |
| CN112805725A (en) * | 2020-01-06 | 2021-05-14 | 深圳市微蓝智能科技有限公司 | Data processing method and device and computer readable storage medium |
| WO2021138783A1 (en) * | 2020-01-06 | 2021-07-15 | 深圳市微蓝智能科技有限公司 | Data processing method and apparatus, and computer readable storage medium |
| CN111309706A (en) * | 2020-01-20 | 2020-06-19 | 北京明略软件系统有限公司 | Model training method and device, readable storage medium and electronic equipment |
| US20210334440A1 (en) * | 2020-04-28 | 2021-10-28 | Mitsubishi Heavy Industries, Ltd. | Model optimization device, model optimization method, and program |
| US20210334702A1 (en) * | 2020-04-28 | 2021-10-28 | Mitsubishi Heavy Industries, Ltd. | Model evaluating device, model evaluating method, and program |
| CN111708810A (en) * | 2020-06-17 | 2020-09-25 | 北京世纪好未来教育科技有限公司 | Model optimization recommendation method and device and computer storage medium |
| CN111708810B (en) * | 2020-06-17 | 2022-05-27 | 北京世纪好未来教育科技有限公司 | Model optimization recommendation method and device and computer storage medium |
| CN112070238A (en) * | 2020-11-10 | 2020-12-11 | 鹏城实验室 | Accurate machine learning asynchronous prediction method and system and storage medium |
| CN112598326A (en) * | 2020-12-31 | 2021-04-02 | 五八有限公司 | Model iteration method and device, electronic equipment and storage medium |
| WO2022213565A1 (en) * | 2021-04-07 | 2022-10-13 | 全球能源互联网研究院有限公司 | Review method and apparatus for prediction result of artificial intelligence model |
| CN113536004A (en) * | 2021-06-25 | 2021-10-22 | 猫岐智能科技（上海）有限公司 | Real-time graph monitoring system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108875963A (en) | Optimization method, device, terminal device and the storage medium of machine learning model | |
| CN106529565B (en) | Model of Target Recognition training and target identification method and device calculate equipment | |
| CN109063742B (en) | Butterfly identification network construction method and device, computer equipment and storage medium | |
| KR102114564B1 (en) | Learning system, learning device, learning method, learning program, teacher data creation device, teacher data creation method, teacher data creation program, terminal device and threshold change device | |
| CN112735535B (en) | Prediction model training method, prediction model training device, data prediction method, data prediction device and storage medium | |
| CN109583489A (en) | Defect classifying identification method, device, computer equipment and storage medium | |
| CN108805160A (en) | Transfer learning method, apparatus, computer equipment and storage medium | |
| CN108256555A (en) | Picture material recognition methods, device and terminal | |
| CN111814385A (en) | Method, device and computer equipment for predicting quality of workpiece | |
| CN111095293A (en) | Image aesthetic processing method and electronic equipment | |
| CN107690660A (en) | Image-recognizing method and device | |
| CN109949290A (en) | Pavement crack detection method, device, equipment and storage medium | |
| WO2020119419A1 (en) | Image recognition-based testing and apparatus, and computer device and storage medium | |
| CN110472594A (en) | Method for tracking target, information insertion method and equipment | |
| CN111507419A (en) | Training method and device of image classification model | |
| CN108846340A (en) | Face identification method, device and disaggregated model training method, device, storage medium and computer equipment | |
| CN107944708A (en) | Borrow or lend money the model discrimination method, apparatus and storage medium of risk control | |
| CN110135582B (en) | Neural network training method, neural network training device, image processing method, image processing device and storage medium | |
| CN106778820A (en) | Identification model determines method and device | |
| CN110942012A (en) | Image feature extraction method, pedestrian re-identification method, device and computer equipment | |
| CN114781272A (en) | Carbon emission prediction method, device, equipment and storage medium | |
| CN112101162A (en) | Image recognition model generation method and device, storage medium and electronic equipment | |
| CN109993021A (en) | The positive face detecting method of face, device and electronic equipment | |
| CN111598213A (en) | Network training method, data identification method, device, equipment and medium | |
| CN107844774A (en) | System of selection, device, intelligent terminal and the storage medium that image is shown |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181123 |