CN110443126A - Model hyper parameter adjusts control method, device, computer equipment and storage medium - Google Patents
Model hyper parameter adjusts control method, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN110443126A CN110443126A CN201910569483.XA CN201910569483A CN110443126A CN 110443126 A CN110443126 A CN 110443126A CN 201910569483 A CN201910569483 A CN 201910569483A CN 110443126 A CN110443126 A CN 110443126A
- Authority
- CN
- China
- Prior art keywords
- information
- model
- tune
- target
- hyper parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Abstract
The embodiment of the invention discloses a kind of model hyper parameter adjustment control method, device, computer equipment and storage mediums, include the following steps: to obtain the raw information to training pattern, wherein, raw information includes the sample information to the model information of training pattern and for training to the sample data of training pattern；Model information and sample information calculate according to preset performance algorithm and generate model performance information；Target tune corresponding with model performance information is searched in preset tune ginseng database according to model performance information and joins scheme, so as to join project setting hyper parameter according to target tune to training pattern.The embodiment of the present invention is by obtaining to the model information of training pattern and the sample information of sample data and calculating generation model performance information, target tune, which is chosen, further according to the model performance information joins scheme, so that joining the adjustment that scheme carries out hyper parameter according to target tune to training pattern, the difficulty of adjustment hyper parameter can be effectively reduced and improving tune ginseng efficiency.
Description
Technical field
The present embodiments relate to model training technical field, especially a kind of model hyper parameter adjustment control method, dress
It sets, computer equipment and storage medium.
Background technique
With the development of science and technology, being related to the mass data of every aspect in the work and life of people, in big data
Under generation, analysis and modeling is carried out based on the mass data in each scene and has become the daily work of major artificial intelligence field
With scientific research content, each class model is many since the complexity of itself can exist cannot be by the ginseng that is learnt to data
Number, this parameter are called hyper parameter.
The value of hyper parameter often will have a direct impact on prediction effect of the established model under different scenes, so modeling
It needs to be adjusted hyper parameter in the process to improve the accuracy of model, still, due to the diversity and super ginseng of hyper parameter
The uncertainty in space, the complexity for causing hyper parameter tune to be joined is very costly and time consuming serious, reduces tune ginseng efficiency.
Summary of the invention
The embodiment of the present invention provides a kind of by taking tune ginseng scheme according to sample and model to improve the model for adjusting ginseng efficiency
Hyper parameter adjusts control method, device, computer equipment and storage medium.
In order to solve the above technical problems, the technical solution that the embodiment of the invention uses is: providing a kind of mould
Type hyper parameter adjusts control method, includes the following steps:
Obtain the raw information to training pattern, wherein the raw information includes that the model to training pattern is believed
It ceases and for the sample information described in training to the sample data of training pattern；
The model information and the sample information calculate according to preset performance algorithm and generate model performance letter
Breath；
It is searched in preset tune ginseng database according to the model performance information corresponding with the model performance information
Target tune join scheme so that it is described to training pattern according to the target tune join project setting hyper parameter.
Optionally, described to be searched and the model performance in preset tune ginseng database according to the model performance information
The corresponding target tune of information joins scheme, so that the step for joining project setting hyper parameter according to the target tune to training pattern
Further include such as following step before rapid:
Obtain preset local terminal performance data list, wherein the performance data list includes the local terminal
Machine configuration information；
Algorithm is divided to carry out calculating generation machine performance information and add the machine configuration information according to the race of preset machine
It adds in the model performance information.
Optionally, further include such as following step before the step of acquisition preset local terminal performance data list:
Obtain the device number information of the local terminal；
Control information is detected according to the device number information generating device, to detect control information to institute according to the equipment
It states local terminal and carries out the detection acquisition machine configuration information.
Optionally, described to be searched and the model performance in preset tune ginseng database according to the model performance information
The corresponding target tune of information joins scheme, so that the step for joining project setting hyper parameter according to the target tune to training pattern
Further include such as following step after rapid:
The model performance information and target tune ginseng scheme are subjected to structuring conversion and generate target storage information；
The target is stored into information preservation into preset tune ginseng database.
Optionally, described that the target is stored into step of the information preservation into preset tune ginseng database including as follows
State step:
It obtains and the identity information to the corresponding model training operator of training pattern；
The flag information of the target storage information is set and is stored to the tune according to the identity information and joins database
In, so that target storage information is associated with the operator.
Optionally, the step of the acquisition and the identity information to the corresponding model training operator of training pattern
Suddenly, including such as following step:
Obtain the facial image of the operator；
The facial image is input in preset human face recognition model, wherein the human face recognition model is training
To convergent convolutional neural networks model；
Obtain the identity information of the operator of the human face recognition model output.
Optionally, described that the target is stored into step of the information preservation into preset tune ginseng database including as follows
State step:
It is established by thread and the target is stored into information preservation to the preset store tasks adjusted in ginseng database；
Detect the operation for being higher than the store tasks in the task queue after the store tasks with the presence or absence of priority
Task；
When the task queue there are priority be higher than the store tasks operation task when, preferentially execute the operation
Readjustment executes the store tasks after task is finished to the operation task.
In order to solve the above technical problems, the embodiment of the present invention also provides a kind of model hyper parameter adjustment control device, comprising:
First obtain module, for obtaining the raw information to training pattern, wherein the raw information include described in
The model information of training pattern and sample information for the training sample data to training pattern；
First processing module, based on being carried out according to preset performance algorithm to the model information and the sample information
It calculates and generates model performance information；
First execution module, for being searched and the mould in preset tune ginseng database according to the model performance information
The corresponding target tune of type performance information joins scheme, so that described join the super ginseng of project setting according to the target tune to training pattern
Number.
Optionally, further includes:
Second obtains module, for obtaining preset local terminal performance data list, wherein the performance data list
Machine configuration information including the local terminal；
Second execution module, for dividing algorithm to carry out calculating generation to the machine configuration information according to the race of preset machine
Machine performance information is simultaneously added in the model performance information.
Optionally, further includes:
Third obtains module, for obtaining the device number information of the local terminal；
Third execution module, for detecting control information according to the device number information generating device, to be set according to
Standby detection control information carries out detection to the local terminal and obtains the machine configuration information.
Optionally, further includes:
Second processing module, for the model performance information and target tune ginseng scheme to be carried out structuring conversion
It generates target and stores information；
4th execution module, for the target to be stored information preservation into preset tune ginseng database.
Optionally, further includes:
First acquisition submodule, for obtaining and the identity to the corresponding model training operator of training pattern
Information；
First implementation sub-module, for the flag information of the target storage information to be arranged according to the identity information and deposits
Storage is into tune ginseng database, so that target storage information is associated with the operator.
Optionally, further includes:
Second acquisition submodule, for obtaining the facial image of the operator；
First processing submodule, for the facial image to be input in preset human face recognition model, wherein described
Human face recognition model is to train to convergent convolutional neural networks model.
Third acquisition submodule, the identity information of the operator for obtaining the human face recognition model output.
Optionally, further includes:
The target is stored information preservation into preset tune ginseng database for establishing by thread by thread submodule
Store tasks；
Detection sub-module, for detect in the task queue after the store tasks be higher than with the presence or absence of priority it is described
The operation task of store tasks；
Second implementation sub-module, for there are the operation tasks that priority is higher than the store tasks when the task queue
When, it preferentially executes readjustment after the operation task is finished to the operation task and executes the store tasks.
In order to solve the above technical problems, the embodiment of the present invention also provides a kind of computer equipment, including memory and processing
Device is stored with computer-readable instruction in the memory, when the computer-readable instruction is executed by the processor, so that
The processor executes the step of above-mentioned model hyper parameter adjustment control method.
In order to solve the above technical problems, the embodiment of the present invention also provides a kind of storage Jie for being stored with computer-readable instruction
Matter, when the computer-readable instruction is executed by one or more processors, so that one or more processors execute above-mentioned mould
Type hyper parameter adjusts the step of control method.
The embodiment of the present invention has the beneficial effect that by obtaining the raw information to training pattern, including the mould to be trained
The model information of type and the sample information of sample data, sample data are the then bases for training this to wait for training pattern
Performance algorithm carries out calculating generation to the sample information of model information and sample data, and this waits for the model performance information of training pattern,
Further according to the model performance information adjust ginseng database in search target tune join scheme so that should to training pattern according to be somebody's turn to do
Target tune joins the adjustment that scheme carries out hyper parameter, by automatic according to the input sample data and model information to training pattern
Change selection and adjust ginseng scheme, the tune ginseng scheme of selection is adapted to training pattern and sample data, and tune ginseng difficulty can be effectively reduced and mention
Ginseng efficiency to a high-profile.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those skilled in the art, without creative efforts, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is the basic procedure schematic diagram that model of embodiment of the present invention hyper parameter adjusts control method；
Fig. 2 is the flow diagram that the embodiment of the present invention obtains terminal capabilities；
Fig. 3 is the flow diagram that the embodiment of the present invention obtains machine configuration information；
Fig. 4 is the flow diagram that the embodiment of the present invention stores that target tune joins scheme；
Fig. 5 is the flow diagram that ginseng scheme and operator are adjusted in association of the embodiment of the present invention；
Fig. 6 is the flow diagram for the identity information that the embodiment of the present invention obtains operator；
Fig. 7 is the flow diagram that multithreading of the embodiment of the present invention stores that target stores information；
Fig. 8 is that model of embodiment of the present invention hyper parameter adjusts control device basic structure schematic diagram；
Fig. 9 is computer equipment of embodiment of the present invention basic structure block diagram.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, below in conjunction in the embodiment of the present invention
Attached drawing, technical scheme in the embodiment of the invention is clearly and completely described.
In some processes of the description in description and claims of this specification and above-mentioned attached drawing, contain according to
Multiple operations that particular order occurs, but it should be clearly understood that these operations can not be what appears in this article suitable according to its
Sequence is executed or is executed parallel, and serial number of operation such as 101,102 etc. is only used for distinguishing each different operation, serial number
It itself does not represent and any executes sequence.In addition, these processes may include more or fewer operations, and these operations can
To execute or execute parallel in order.It should be noted that the description such as " first " herein, " second ", is for distinguishing not
Same message, equipment, module etc., does not represent sequencing, does not also limit " first " and " second " and be different type.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those skilled in the art's every other implementation obtained without creative efforts
Example, shall fall within the protection scope of the present invention.
Embodiment 1
It is the basic procedure schematic diagram that the present embodiment model hyper parameter adjusts control method referring specifically to Fig. 1, Fig. 1.
As shown in Figure 1, a kind of model hyper parameter adjusts control method, include the following steps:
The raw information of S1100, acquisition to training pattern, wherein the raw information includes described to training pattern
Model information and sample information for the training sample data to training pattern；
Refer to the model analyzed and established according to the mass data of application scenarios to training pattern, to the original of training pattern
Information includes the model information and sample information that wait for training pattern, wherein sample information is should mould be trained for training
The sample data of type, when implementing, the raw information to training pattern can be inputted by user, with model hyper parameter tune of the present invention
Whole control method is applied to for user terminal, and user terminal includes but is not limited to smart phone, Intelligent bracelet, plate, PC
(personal computer, personal computer) terminal and other intelligent electronic devices, such as pass through display in PC terminal
Show raw information input interface to training pattern to user, system obtained by the operation and input of monitoring users this to
The raw information of training pattern；On the other hand, it should can also detect to obtain by user terminal to the raw information of training pattern, example
Such as during modeling, the selected model of system detection and sample data respectively obtain model information and sample information, wherein
Sample data is to preset to be input in system, which includes the title and model attributes information to training pattern,
Sample information includes the data size information of sample data.
S1200, the model information and the sample information are carried out according to preset performance algorithm to calculate generation model
Performance information；
After obtaining to the model information of training pattern and the sample information of sample data, system is according to preset property
Energy algorithm calculates the model information and sample information to generate model performance information, wherein performance algorithm is in system
The preset algorithm for calculating generation model performance information according to training sample data and model can basis when implementing
The size computation model grade of model information and sample data generates model performance information, such as: according to the size of sample data
It is divided into first order sample, second level sample and third pole sample from high to low, wherein the data of the sample data of first order sample
Amount is less than 1G (Giga, giga-), and the data volume of the sample data of second level sample is greater than 1G and is less than 5G, the sample of third level sample
The data volume of notebook data is greater than 5G, and system, which calculate according to the size of model information and sample data generating, carries tune ginseng etc.
The model performance information of grade, such as when sample data is first order sample and model is machine learning model, the tune of generation is joined
Grade is that first order tune joins grade, and when sample data is second level sample and model is machine learning model, the tune of generation is joined
Grade is that second level tune joins grade, and when sample data is third level sample and model is machine learning model, the tune of generation is joined
Grade is that third level tune joins grade, can generate model performance information according to the sample information of model information and sample data is corresponding.
S1300, it is searched and the model performance information in preset tune ginseng database according to the model performance information
Corresponding target tune joins scheme, so that described join project setting hyper parameter according to the target tune to training pattern.
After calculating generation model performance information, system chooses target in adjusting ginseng database according to the model performance information
Ginseng scheme is adjusted, so that the adjustment that scheme carries out hyper parameter should be joined according to the target tune to training pattern, when implementing, adjusts parameter evidence
Library is the preset warehouse for storage and management tune ginseng scheme, wherein adjusting ginseng scheme is that ginseng is adjusted in preset hyper parameter automation
Mechanism, such as the multifunction automatic tune that can be integrated according to BayesianOptimization and GridSearch join mechanism,
BayesianOptimization refers to that Bayes optimizes, and so-called optimization, actually one is asked the process of extreme value, data section
What is learned is many times exactly the problem of seeking extreme value.Such as differentiating is exactly the good method for seeking extreme value, the item of the optimization based on gradient
Part be derivative can be just found out known to functional form, and if function convex function just can be with, however be actually many times not
Meet the two conditions, so gradient optimizing cannot be used, Bayes's optimization is come into being, and Bayes's optimization is usually used in solving
Inversion problem, inversion problem refer to the parameter for leaving for determining characterization problems feature by result and certain General Principles (or model)
Or model parameter.Grid Search (exhaustive search, grid search) is a kind of tune ginseng means, in the parameter selection of all candidates
In, by looping through, each possibility is attempted, the parameter to behave oneself best is exactly final result.Its principle be like
Maximum value is looked in array.(why it is grid search by taking the model there are two parameter as an example, parameter a has 3 kinds of possibility, parameter b
There are 4 kinds of possibility, all possibilities are listed, the grid of a 3*4 can be expressed as, cyclic process is like in each grid
In traverse and search).
When implementing, model hyper parameter adjustment control method of the present invention is based on input sample amount (data volume of sample data)
And corresponding model (to training pattern) carries out automation matching and adjusts ginseng scheme, and such as: when the data volume of sample data is less than 1G
And join scheme to training pattern for the target tune that machine learning model then selects Grid Search integrated, when the number of sample data
It is greater than 1G according to amount and is the target tune ginseng that machine learning model then selects BayesianOptimization integrated to training pattern
Scheme.
In another embodiment, it can be combined with machine performance and choose target tune ginseng scheme, machine performance refers to execution
The performance of the equipment of model training, such as according to the sample information, model information and machine performance of sample data (such as machine
Configuration, whether there is or not available GPU etc.) automation matching adjust ginseng scheme, by the data volume of sample data be less than 1G for, when model is
Machine learning model and the target tune ginseng scheme for thering is available GPU then to select Grid Search integrated, when model is machine learning mould
The type and target tune without can be used GPU then to select BayesianOptimization integrated joins scheme, by by Bayes with
The tune such as Grid Search ginseng entirety carries out automation and integrates, by according to input sample amount, machine performance and selected mould
Type automation carries out optimal tune and joins Scheme Choice, to choose tune ginseng side compatible with model, data volume and machine performance
Case.In one embodiment, machine performance can also be calculated according to the configuration of machine, such as run component software by machine
(such as Shandong great master, 3DMark, Furmark or SiSoftware Sandra Lite) is detected and analyzed equipment, example
As the data volume of sample data is less than 1G, model is machine learning model and is detected by Shandong great master race component software
Score very much less than 15 when, the target tune for selecting Grid Search integrated joins scheme, passes through the size class according to sample data
Not, model information and the compatible tune ginseng scheme of combination machine automatic performance selection, substantially increase tune ginseng efficiency.
The present embodiment waits for the model information and sample of training pattern including this by obtaining the raw information to training pattern
The sample information of notebook data, sample data are for training this to wait for training pattern, then according to performance algorithm to model information
Carrying out calculating generation with the sample information of sample data, this waits for the model performance information of training pattern, believes further according to the model performance
Breath searches target tune in adjusting ginseng database and joins scheme, is surpassed so that should join scheme according to the target tune to training pattern
The adjustment of parameter, by adjusting ginseng scheme, choosing according to input sample data and model information the automation selection to training pattern
The tune ginseng scheme taken is adapted to training pattern and sample data, and tune ginseng difficulty can be effectively reduced and improve tune ginseng efficiency.
In one alternate embodiment, referring to Fig. 2, Fig. 2 is that one embodiment of the invention obtains the specific of terminal capabilities
Flow diagram.
As shown in Fig. 2, further including such as following step before step 1100:
S1010, preset local terminal performance data list is obtained, wherein the performance data list includes described
The machine configuration information of ground terminal；
The hardware configuration information tables of data that performance data list is local terminal is obtained, when implementing, ginseng super for model
Several automatic configuration needs are completed by local terminal, then system, which needs to be deployed according to the hardware configuration of local terminal, adjusts ginseng side
Case.Local terminal includes but is not limited to smart phone, tablet computer, PC (personal computer, personal computer) terminal
And other electronic equipments, local terminal run the support for needing hardware components, the performance number that system passes through acquisition local terminal
According to list, the machine configuration information of local terminal can be obtained according to the performance data list.In one embodiment, local whole
It is provided with database in end, performance data list is stored in the database, system can obtain this by accessing the database
The performance data list of ground terminal, the machine configuration information in the performance data list can be inputted by user oneself to be saved, example
If local terminal is provided with display, system is by display to user's display performance data list input interface, and user is at this
The machine configuration information of local terminal is inputted in input interface, the operation of system monitoring users stores the machine configuration information
Into performance data list, then performance data list is stored into database.
S1020, algorithm is divided to carry out calculating generation machine performance letter to the machine configuration information according to the race of preset machine
It ceases and is added in the model performance information.
After the performance data list for obtaining local terminal, system divides algorithm to arrange performance data according to the race of preset machine
Machine configuration information in table, which calculate, generates machine performance information, and machine performance information characterizes the machine of local terminal
Can, and the machine performance information is added in model performance information, system can be according to input sample amount (to training pattern
Sample data), machine performance and carry out optimal tunes to training pattern automation and join Scheme Choice, to choose and model, number
According to amount and the compatible tune ginseng scheme of machine performance.
In one embodiment, machine, which is run, divides algorithm to be calculating locally eventually for being configured according to terminal hardware for systemic presupposition
The tool of the machine performance at end, such as component software is run by machine, equipment is detected and analyzed, running component software includes but not
It is limited to Lu great master, 3DMark, Furmark and SiSoftware Sandra Lite, the present embodiment passes through according to sample data
Size rank, model information and the machine performance automation of local terminal is combined to select compatible tune ginseng scheme, can be effective
It improves and adjusts ginseng efficiency.
In another alternative embodiment, referring to Fig. 3, Fig. 3 is that one embodiment of the invention obtains machine configuration information
Basic procedure schematic diagram.
As shown in figure 3, further including such as following step before step 1010:
S1001, the device number information for obtaining the local terminal；
Device number information is the sequence number of local terminal, and when implementing, the device number information of local terminal can be by user
It is configured, so that local terminal is distinguished, such as: the device number information of local terminal is device number for " ZD001 "
With the database of local terminal, system can obtain setting for local terminal by the database of access local terminal for information storage
Standby number information.
S1002, control information is detected according to the device number information generating device, to detect control letter according to the equipment
Breath carries out detection to the local terminal and obtains the machine configuration information.
After the device number information for obtaining local terminal, system detects control letter according to the device number information generating device
Breath, and control information is detected according to the equipment and carries out the machine configuration information for detecting to obtain local terminal to local terminal,
When implementing, equipment detection control information can control local terminal operation, and detection is local eventually in the operational process of local terminal
The machine at end configures, such as system detects the preset inspection software (such as Shandong great master) of control information control by equipment and detects this
The machine of ground terminal configures, and can accurately obtain the machine configuration information of local terminal.
In one alternate embodiment, referring to Fig. 4, Fig. 4 is one embodiment of the invention storage target tune ginseng scheme
Basic procedure schematic diagram.
As shown in figure 4, further including such as following step after step 1300:
S1400, the model performance information and target tune ginseng scheme progress structuring conversion are generated target and deposited
Store up information；
Target tune ginseng scheme is chosen again and project setting is joined after the hyper parameter of training pattern according to the target tune, and system will
The model performance information and target tune ginseng scheme carry out structuring processing, so that target storage information is generated, target storage letter
Breath uses structured storage mode, and so-called structured storage method is really applied to the principle of tree file system individually
File in so that single file also can include " subdirectory " as file system, " subdirectory " can also include deeper
Secondary " subdirectory ", each " subdirectory " can contain multiple files, the content for needing multiple files to store originally by tree-shaped knot
Structure and level are saved in a file.System carries out structuring turn by the way that model performance information and target tune are joined scheme
The target storage information of key-value pair form is changed into, to establish model performance and adjust contacting between ginseng scheme.
S1500, the target is stored to information preservation into preset tune ginseng database.
After model performance information and target tune ginseng scenario-frame are converted into target storage information by system, which is deposited
Storage information is stored to adjusting in ginseng database, and tune ginseng database is systemic presupposition for storage and management target storage result
Warehouse, the present embodiment are stored by joining model performance information and target tune after scheme forms key-value pair into tune ginseng database,
To ginseng scheme be adjusted to associate with model performance, the later period is facilitated to adjust ginseng for calling to training pattern for same model performance
Scheme does not need to repeat to calculating to training pattern with same model performance, and simplified model tune joins step, improves mould
Type training effectiveness.
In one alternate embodiment, referring to Fig. 5, Fig. 5 is that ginseng scheme and operation are adjusted in one embodiment of the invention association
The basic procedure schematic diagram of personnel.
As shown in figure 5, step 1500 includes such as following step:
S1510, it obtains and the identity information to the corresponding model training operator of training pattern；
Identity information is the proof of identification information of operator, including but not limited to the name, work number of operator, department
And the information such as ID card No., when implementing, the identity information of operator can be by operator oneself operation input, example
If model hyper parameter of the present invention adjustment control method is applied on smart machine, which includes display, and system passes through
The display shows identity information input interface to operator, and operator inputs correspondence in the identity information input interface
Identity information, such as in the identity information input interface include name input field, work number input field and ID card No. input
Column, operator can input respectively name, work number in the name input field, work number input field and ID card No. input field
And ID card No., the input of system snoop-operations personnel are operated to obtain the identity information of operator.It is, of course, also possible to adopt
The identity information of operator is obtained with voice mode, such as acquires the voice messaging of operator's input, in the voice messaging
Name information and ID card No. information including operator, system obtain the identity of operator by speech recognition technology
Information, speech recognition technology are exactly to allow machine that voice signal is changed into corresponding text or order by identification and understanding process
Technology.
S1520, the flag information of the target storage information is set according to the identity information and is stored to the tune and is joined
In database, so that target storage information is associated with the operator.
After the identity information for obtaining operator, the mark letter that target stores information is arranged according to the identity information for system
Breath, then by target storage information preservation to adjusting in ginseng database, when implementing, flag information is the mark of target storage information
Sign information, label be for famous special product target (target storage information) and classify or content (identity information of operator),
To establish being associated with for target storage information and operator, the later period is facilitated to find the responsible person of model training.
In one alternate embodiment, referring to Fig. 6, Fig. 6 is the identity that one embodiment of the invention obtains operator
The basic procedure schematic diagram of information.
As shown in fig. 6, step S1510 includes such as following step:
S1511, the facial image for obtaining the operator；
Facial image refers to the Facial Expression Image of operator, when implementing, is adjusted and is controlled with model hyper parameter of the present invention
Method processed is applied to for local terminal, and local terminal is provided with for camera, and system acquires operator by camera
Facial image, it is, of course, also possible to acquire the facial image of operator by user terminal, user terminal includes but is not limited to
Smart phone, Intelligent bracelet, tablet computer and other electronic equipments for being provided with camera, such as the preposition of smart phone are taken the photograph
As head or rear camera, operator acquires the Facial Expression Image of oneself by the camera of smart phone, and uploads
To local terminal.When implementing, the facial image of operator can also be obtained by way of taking pictures or shooting the video, with logical
It crosses for the facial image that the mode to shoot the video obtains operator, system is shot to obtain by camera to operator
Target video, system can handle target video by Video processing software (such as OpenCV), target video torn open
It is divided into several frame pictures, picture image is extracted from target video by timing acquiring mode.Such as with 0.5 second one speed
A Target Photo is extracted in target video, then randomly selects a target figure again in obtained several Target Photos
Facial image of the piece as operator；But it is not limited to this, according to the difference of concrete application scene, acquire picture image
Speed is able to carry out the adjustment of adaptability, and Adjustment principle is, system processing capacity is stronger and tracks accuracy requirement more Gao Ze
Acquisition time is shorter, reach with picture pick-up device acquisition image Frequency Synchronization when until；Otherwise, then acquisition time interval is longer,
But longest acquisition time interval must not exceed 1s.It is of course also possible to directly be randomly selected in several frame pictures of target video
Facial image of one picture as operator.
S1512, the facial image is input in preset human face recognition model, wherein the human face recognition model
For training to convergent convolutional neural networks model；
Human face recognition model is the tool of the pre-set facial image for identification of system, when implementing, be can be used
LSTM network (shot and long term remembers artificial nerve network model, Long Short-Term Memory) is used as neural network model.
LSTM network is controlled discarding by " door " (gate) or increases information, to realize the function of forgetting or memory." door " is
A kind of structure passing through header length is operated and is formed by a sigmoid (S sigmoid growth curve) function and a dot product.
The output valve of sigmoid function represents discarding completely in [0,1] section, 0, and 1 representative passes through completely.It trains to convergent nerve net
Network model has the classifier that can identify facial image, wherein human face recognition model includes above-mentioned neural network model, should
Neural network model includes N+1 classifier, and N is positive integer.
Specifically, by the way that facial image to be input in preset human face recognition model, facial image is obtained in classifier
In classification results, wherein classification results include the confidence of the classification of facial image corresponding identity information and identity information classification
It spends (Confidence).Wherein, the confidence level of identity information classification refers to that facial image is screened by human face recognition model
After classification, facial image, which is classified into more than one identity information classification and obtains facial image, accounts for identity information classification
Percentage value.Due to finally obtaining the corresponding identity information of facial image as one kind, therefore need each of same facial image
The confidence level of identity information classification is compared, for example, system acquisition is classified into Zhang San's to the facial image of operator
Confidence level is 0.95, and the confidence level for being classified into Li Si is 0.63.
Then the confidence level is compared with preset first threshold, when the confidence level is greater than preset first threshold
When, confirm the identity information classification results that the confidence level is characterized for the identity information of operator.Preset first threshold
The numerical value being traditionally arranged to be between 0.9 to 1.By filtering out emotional information of the confidence level greater than first threshold as final body
The identity information that part information classification results, i.e. confirmation confidence level are characterized.For example, when preset first threshold is 0.9, and
The confidence level that the facial image of operator is classified into Zhang San is 0.95, due to 0.95 > 0.9, so the facial image institute table
The identity information of sign is the personally identifiable information of Zhang San.
S1513, the identity information for obtaining the operator that the human face recognition model exports.
Human face recognition model exports the personally identifiable information of operator after identifying to the facial image of operator,
By the way that facial image to be input in preset human face recognition model, and obtain the body of the facial image of human face recognition model output
The confidence level of part information classification, when confidence level is greater than preset first threshold value, the identity information that confirmation confidence level is characterized is classified
As a result it is the personally identifiable information of operator, to improve the identity information classification accuracy of identification facial image.In reality
Shi Shi, when human face recognition model recognizes the name of operator according to the facial image of operator, acceptable basis should
Name searches the personal information of operator corresponding with the name, of the operator in preset customer data base
People's information includes name, identification card number and work number of operator etc..
In one alternate embodiment, referring to Fig. 7, Fig. 7 is the storage target storage of one embodiment of the invention multithreading
The basic procedure schematic diagram of information.
As shown in fig. 7, step S1500 includes such as following step:
S1530, it is established by thread and adjusts the storage in ginseng database to appoint to preset target storage information preservation
Business；
Thread is a single sequential control process in application program.Have in process one it is relatively independent, schedulable
Execution unit, be system independently dispatch and assign CPU basic unit instruction operation when program thread.Single
Running multiple threads completes different work, referred to as multithreading simultaneously in program.Believed by establishing to execute to store the target
Breath is saved to the preset pending task adjusted in ginseng database, so that target storage information preservation to preset tune be joined
The asynchronous multithreading of operation task of operation and other operation tasks and other application program in database carries out simultaneously.
It is higher than the store tasks with the presence or absence of priority in task queue after S1540, the detection store tasks
Operation task；
Task queue refers to including multiple operation tasks, and asynchronous call is carried out between these operation tasks, to solve
The set of tasks of task blocking problem, the operation task in task queue are provided with corresponding priority, priority
(priority) it is a kind of agreement, is computer time sharing system when handling multiple operation procedures, determines each operation journey
Sequence receives the parameter of the priority level of system resource, and priority is high first to be done, and does after priority is low.System passes through traversal task
Each operation task in queue is compared with the priority of pending task, to search in task queue with the presence or absence of excellent
First grade is higher than the operation task of pending task.
S1550, when the task queue there are priority be higher than the store tasks operation task when, preferentially execute institute
It states readjustment after operation task is finished to the operation task and executes the store tasks.
Preferential execution priority is higher than other operation tasks of the pending task, and system operation smoothness can be made not block
, such as to handle multiple training patterns in same time system and store the operation of target storage information, system first carries out instruction
The pending task is executed again after practicing the task of model, improves model training efficiency.
In order to solve the above technical problems, the embodiment of the present invention also provides a kind of model hyper parameter adjustment control device.
It is that the present embodiment model hyper parameter adjusts control device basic structure schematic diagram referring specifically to Fig. 8, Fig. 8.
As shown in figure 8, a kind of model hyper parameter adjusts control device, comprising: first, which obtains module 2100, first, handles mould
Block 2200 and the first execution module 2300, wherein the first acquisition module 2100 is used to obtain the raw information to training pattern,
In, the raw information includes the model information to training pattern and for the training sample number to training pattern
According to sample information；First processing module 2200 is used to believe the model information and the sample according to preset performance algorithm
Breath, which calculate, generates model performance information；First execution module 2300 is used for according to the model performance information in preset tune
Join and search corresponding with model performance information target tune ginseng scheme in database so that it is described to training pattern according to institute
State target tune ginseng project setting hyper parameter.
The present embodiment waits for the model information and sample of training pattern including this by obtaining the raw information to training pattern
The sample information of notebook data, sample data are for training this to wait for training pattern, then according to performance algorithm to model information
Carrying out calculating generation with the sample information of sample data, this waits for the model performance information of training pattern, believes further according to the model performance
Breath searches target tune in adjusting ginseng database and joins scheme, is surpassed so that should join scheme according to the target tune to training pattern
The adjustment of parameter, by adjusting ginseng scheme, choosing according to input sample data and model information the automation selection to training pattern
The tune ginseng scheme taken is adapted to training pattern and sample data, and tune ginseng difficulty can be effectively reduced and improve tune ginseng efficiency.
In some embodiments, model hyper parameter adjusts control device further include: second, which obtains module and second, executes
Module, wherein the second acquisition module is for obtaining preset local terminal performance data list, wherein the performance data column
Table includes the machine configuration information of the local terminal；Second execution module is used to divide algorithm to described according to the race of preset machine
Machine configuration information, which calculate, to be generated machine performance information and is added in the model performance information.
In some embodiments, model hyper parameter adjusts control device further include: third obtains module and third executes
Module, wherein third obtains the device number information that module is used to obtain the local terminal；Third execution module is used for according to institute
Device number information generating device detection control information is stated, the local terminal is carried out with detecting control information according to the equipment
Detection obtains the machine configuration information.
In some embodiments, model hyper parameter adjusts control device further include: Second processing module and the 4th executes
Module, wherein Second processing module is used to that the model performance information and target tune ginseng scheme to be carried out structuring and be turned
It changes and generates target storage information；4th execution module, which is used to storing the target into information preservation to preset tune, joins database
In.
In some embodiments, model hyper parameter adjusts control device further include: the first acquisition submodule and first is held
Row submodule, wherein the first acquisition submodule is for obtaining with described to the corresponding model training operator of training pattern
Identity information；First implementation sub-module is used to be arranged according to the identity information flag information of the target storage information simultaneously
It stores into tune ginseng database, so that target storage information is associated with the operator.
In some embodiments, model hyper parameter adjusts control device further include: the second acquisition submodule, the first processing
Submodule and third acquisition submodule, wherein the second acquisition submodule is used to obtain the facial image of the operator；First
Processing submodule is for the facial image to be input in preset human face recognition model, wherein the human face recognition model
For training to convergent convolutional neural networks model.Third acquisition submodule is used to obtain the institute of the human face recognition model output
State the identity information of operator.
In some embodiments, model hyper parameter adjust control device further include: thread submodule, detection sub-module and
Second implementation sub-module, wherein thread submodule is used to established by thread by target storage information preservation to preset
Adjust the store tasks in ginseng database；Detection sub-module in the task queue after detecting the store tasks for whether there is
Priority is higher than the operation task of the store tasks；Second implementation sub-module is used for when there are priority height for the task queue
When the operation task of the store tasks, preferentially executes to adjust back after the operation task is finished to the operation task and hold
The row store tasks
About the device in above-described embodiment, wherein modules execute the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
In order to solve the above technical problems, the embodiment of the present invention also provides computer equipment.It is this referring specifically to Fig. 9, Fig. 9
Embodiment computer equipment basic structure block diagram.
As shown in figure 9, the schematic diagram of internal structure of computer equipment.As shown in figure 9, the computer equipment includes passing through to be
Processor, non-volatile memory medium, memory and the network interface of bus of uniting connection.Wherein, the computer equipment is non-easy
The property lost storage medium is stored with operating system, database and computer-readable instruction, can be stored with control information sequence in database
Column when the computer-readable instruction is executed by processor, may make processor to realize a kind of model hyper parameter adjustment control method.
The processor of the computer equipment supports the operation of entire computer equipment for providing calculating and control ability.The computer
It can be stored with computer-readable instruction in the memory of equipment, when which is executed by processor, may make place
Reason device executes a kind of model hyper parameter adjustment control method.The network interface of the computer equipment is used for and terminal connection communication.
It will be understood by those skilled in the art that structure shown in figure, the only block diagram of part-structure relevant to application scheme,
The restriction for the computer equipment being applied thereon to application scheme is not constituted, specific computer equipment may include ratio
More or fewer components as shown in the figure perhaps combine certain components or with different component layouts.
Processor obtains module 2100, first processing module 2200 and for executing in Fig. 8 first in present embodiment
One execution module 2300, program code and Various types of data needed for memory is stored with the above-mentioned module of execution.Network interface is used for
To the data transmission between user terminal or server.Memory in present embodiment is stored with the adjustment control of model hyper parameter
Program code needed for executing all submodules in device and data, server are capable of the program code and data of invoking server
Execute the function of all submodules.
Computer waits for the model information and sample of training pattern including this by obtaining the raw information to training pattern
The sample information of data, sample data be for training this to wait for training pattern, then according to performance algorithm to model information and
The sample information of sample data carries out calculating generation, and this waits for the model performance information of training pattern, further according to the model performance information
Target tune is searched in adjusting ginseng database and joins scheme, so that should join scheme according to the target tune to training pattern carries out super ginseng
Several adjustment is chosen by adjusting ginseng scheme according to input sample data and model information the automation selection to training pattern
Tune ginseng scheme be adapted to training pattern and sample data, can be effectively reduced tune ginseng difficulty improve tune join efficiency.
The present invention also provides a kind of storage mediums for being stored with computer-readable instruction, and the computer-readable instruction is by one
When a or multiple processors execute, so that one or more processors execute the adjustment of model hyper parameter described in any of the above-described embodiment
The step of control method.
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, which can be stored in computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, storage medium above-mentioned can be magnetic
The non-volatile memory mediums such as dish, CD, read-only memory (Read-Only Memory, ROM) or random storage memory
Body (Random Access Memory, RAM) etc..
It should be understood that although each step in the flow chart of attached drawing is successively shown according to the instruction of arrow,
These steps are not that the inevitable sequence according to arrow instruction successively executes.Unless expressly stating otherwise herein, these steps
Execution there is no stringent sequences to limit, can execute in the other order.Moreover, at least one in the flow chart of attached drawing
Part steps may include that perhaps these sub-steps of multiple stages or stage are not necessarily in synchronization to multiple sub-steps
Completion is executed, but can be executed at different times, execution sequence, which is also not necessarily, successively to be carried out, but can be with other
At least part of the sub-step or stage of step or other steps executes in turn or alternately.
The above is only some embodiments of the invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of model hyper parameter adjusts control method, which is characterized in that include the following steps:
Obtain raw information to training pattern, wherein the raw information include the model information to training pattern with
And the sample information for the training sample data to training pattern；
The model information and the sample information calculate according to preset performance algorithm and generate model performance information；
Mesh corresponding with the model performance information is searched in preset tune ginseng database according to the model performance information
Mark adjusts ginseng scheme, so that described join project setting hyper parameter according to the target tune to training pattern.
2. model hyper parameter according to claim 1 adjusts control method, which is characterized in that described according to the model
Energy information searches target tune corresponding with the model performance information in preset tune ginseng database and joins scheme, so that described
Further include such as following step before the step of joining project setting hyper parameter according to the target tune to training pattern:
Obtain preset local terminal performance data list, wherein the performance data list includes the machine of the local terminal
Device configuration information；
Algorithm is divided to carry out calculating generation machine performance information and be added to the machine configuration information according to the race of preset machine
In the model performance information.
3. model hyper parameter according to claim 2 adjusts control method, which is characterized in that described to obtain preset local
Further include such as following step before the step of terminal capabilities data list:
Obtain the device number information of the local terminal；
Control information is detected according to the device number information generating device, to detect control information to described according to the equipment
Ground terminal carries out detection and obtains the machine configuration information.
4. model hyper parameter according to claim 1 adjusts control method, which is characterized in that described according to the model
Energy information searches target tune corresponding with the model performance information in preset tune ginseng database and joins scheme, so that described
Further include such as following step after the step of joining project setting hyper parameter according to the target tune to training pattern:
The model performance information and target tune ginseng scheme are subjected to structuring conversion and generate target storage information；
The target is stored into information preservation into preset tune ginseng database.
5. model hyper parameter according to claim 4 adjusts control method, which is characterized in that described to store the target
Step of the information preservation into preset tune ginseng database, including such as following step:
It obtains and the identity information to the corresponding model training operator of training pattern；
The flag information of the target storage information is set according to the identity information and is stored into tune ginseng database, with
It is associated with the target storage information with the operator.
6. model hyper parameter according to claim 5 adjusts control method, which is characterized in that the acquisition is with described wait instruct
The step of practicing the identity information of the corresponding model training operator of model, including such as following step:
Obtain the facial image of the operator；
The facial image is input in preset human face recognition model, wherein the human face recognition model is that training is extremely received
The convolutional neural networks model held back；
Obtain the identity information of the operator of the human face recognition model output.
7. model hyper parameter according to claim 4 adjusts control method, it is characterised in that: described to store the target
Step of the information preservation into preset tune ginseng database, including such as following step:
It is established by thread and the target is stored into information preservation to the preset store tasks adjusted in ginseng database；
Detect the operation task for being higher than the store tasks in the task queue after the store tasks with the presence or absence of priority；
When the task queue there are priority be higher than the store tasks operation task when, preferentially execute the operation task
Readjustment executes the store tasks after being finished to the operation task.
8. a kind of model hyper parameter adjusts control device characterized by comprising
First obtains module, for obtaining the raw information to training pattern, wherein the raw information includes described wait train
The model information of model and sample information for the training sample data to training pattern；
First processing module, for carrying out calculating life to the model information and the sample information according to preset performance algorithm
At model performance information；
First execution module, for being searched and the model in preset tune ginseng database according to the model performance information
The corresponding target tune of energy information joins scheme, so that described join project setting hyper parameter according to the target tune to training pattern.
9. a kind of computer equipment, including memory and processor, it is stored with computer-readable instruction in the memory, it is described
When computer-readable instruction is executed by the processor, so that the processor executes such as any one of claims 1 to 7 right
It is required that the step of model hyper parameter adjustment control method.
10. a kind of storage medium for being stored with computer-readable instruction, the computer-readable instruction is handled by one or more
When device executes, so that one or more processors execute the model hyper parameter as described in any one of claims 1 to 7 claim
The step of adjusting control method.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910569483.XA CN110443126A (en) | 2019-06-27 | 2019-06-27 | Model hyper parameter adjusts control method, device, computer equipment and storage medium |
| PCT/CN2019/103660 WO2020258508A1 (en) | 2019-06-27 | 2019-08-30 | Model hyper-parameter adjustment and control method and apparatus, computer device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910569483.XA CN110443126A (en) | 2019-06-27 | 2019-06-27 | Model hyper parameter adjusts control method, device, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110443126A true CN110443126A (en) | 2019-11-12 |
Family
ID=68428408
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910569483.XA Pending CN110443126A (en) | 2019-06-27 | 2019-06-27 | Model hyper parameter adjusts control method, device, computer equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110443126A (en) |
| WO (1) | WO2020258508A1 (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111696517A (en) * | 2020-05-28 | 2020-09-22 | 平安科技（深圳）有限公司 | Speech synthesis method, speech synthesis device, computer equipment and computer readable storage medium |
| CN112232294A (en) * | 2020-11-09 | 2021-01-15 | 北京爱笔科技有限公司 | Hyper-parameter optimization, target recognition model training and target recognition method and device |
| CN113555008A (en) * | 2020-04-17 | 2021-10-26 | 阿里巴巴集团控股有限公司 | Parameter adjusting method and device for model |
| CN115470910A (en) * | 2022-10-20 | 2022-12-13 | 晞德软件(北京)有限公司 | Automatic parameter adjusting method based on Bayesian optimization and K-center sampling |
| CN115859693A (en) * | 2023-02-17 | 2023-03-28 | 阿里巴巴达摩院(杭州)科技有限公司 | Data processing method and device |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112990480A (en) * | 2021-03-10 | 2021-06-18 | 北京嘀嘀无限科技发展有限公司 | Method and device for building model, electronic equipment and storage medium |
| CN113277388B (en) * | 2021-04-02 | 2023-05-02 | 东南大学 | Data acquisition control method for electric hanging basket |
| CN113822322A (en) * | 2021-07-15 | 2021-12-21 | 腾讯科技（深圳）有限公司 | Image processing model training method and text processing model training method |
| CN113708403A (en) * | 2021-08-17 | 2021-11-26 | 苏州罗约科技有限公司 | Converter grid-connected control method, system, server and storage medium |
| CN114238269B (en) * | 2021-12-03 | 2024-01-23 | 中兴通讯股份有限公司 | Database parameter adjustment method and device, electronic equipment and storage medium |
| CN114392126B (en) * | 2022-01-24 | 2023-09-22 | 佳木斯大学 | Disabled children's hand cooperation training system |
| CN116701350B (en) * | 2023-05-19 | 2024-03-29 | 阿里云计算有限公司 | Automatic optimization method, training method and device, and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764455A (en) * | 2018-05-17 | 2018-11-06 | 南京中兴软件有限责任公司 | Parameter adjustment method, device and storage medium |
| CN109213805A (en) * | 2018-09-07 | 2019-01-15 | 东软集团股份有限公司 | A kind of method and device of implementation model optimization |
| CN109389143A (en) * | 2018-06-19 | 2019-02-26 | 北京九章云极科技有限公司 | A kind of Data Analysis Services system and method for automatic modeling |
| US20190095785A1 (en) * | 2017-09-26 | 2019-03-28 | Amazon Technologies, Inc. | Dynamic tuning of training parameters for machine learning algorithms |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7508961B2 (en) * | 2003-03-12 | 2009-03-24 | Eastman Kodak Company | Method and system for face detection in digital images |
| US20090238426A1 (en) * | 2008-03-19 | 2009-09-24 | Uti Limited Partnership | System and Methods for Identifying an Object within a Complex Environment |
| CN103440493A (en) * | 2013-02-27 | 2013-12-11 | 中国人民解放军空军装备研究院侦察情报装备研究所 | Hyperspectral image blur classification method and device based on related vector machine |
| CN109816000A (en) * | 2019-01-09 | 2019-05-28 | 浙江工业大学 | A kind of new feature selecting and parameter optimization method |
-
2019
- 2019-06-27 CN CN201910569483.XA patent/CN110443126A/en active Pending
- 2019-08-30 WO PCT/CN2019/103660 patent/WO2020258508A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190095785A1 (en) * | 2017-09-26 | 2019-03-28 | Amazon Technologies, Inc. | Dynamic tuning of training parameters for machine learning algorithms |
| CN108764455A (en) * | 2018-05-17 | 2018-11-06 | 南京中兴软件有限责任公司 | Parameter adjustment method, device and storage medium |
| CN109389143A (en) * | 2018-06-19 | 2019-02-26 | 北京九章云极科技有限公司 | A kind of Data Analysis Services system and method for automatic modeling |
| CN109213805A (en) * | 2018-09-07 | 2019-01-15 | 东软集团股份有限公司 | A kind of method and device of implementation model optimization |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113555008A (en) * | 2020-04-17 | 2021-10-26 | 阿里巴巴集团控股有限公司 | Parameter adjusting method and device for model |
| CN111696517A (en) * | 2020-05-28 | 2020-09-22 | 平安科技（深圳）有限公司 | Speech synthesis method, speech synthesis device, computer equipment and computer readable storage medium |
| CN112232294A (en) * | 2020-11-09 | 2021-01-15 | 北京爱笔科技有限公司 | Hyper-parameter optimization, target recognition model training and target recognition method and device |
| CN112232294B (en) * | 2020-11-09 | 2023-10-13 | 北京爱笔科技有限公司 | Super-parameter optimization, target recognition model training and target recognition method and device |
| CN115470910A (en) * | 2022-10-20 | 2022-12-13 | 晞德软件(北京)有限公司 | Automatic parameter adjusting method based on Bayesian optimization and K-center sampling |
| CN115859693A (en) * | 2023-02-17 | 2023-03-28 | 阿里巴巴达摩院(杭州)科技有限公司 | Data processing method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020258508A1 (en) | 2020-12-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443126A (en) | Model hyper parameter adjusts control method, device, computer equipment and storage medium | |
| CN100545856C (en) | Video content analysis system | |
| US10410679B2 (en) | Producing video bits for space time video summary | |
| CN110008875B (en) | Face recognition video clip screening method and system based on key frame backtracking | |
| WO2013170587A1 (en) | Multimedia question and answer system and method | |
| US20110211736A1 (en) | Ranking Based on Facial Image Analysis | |
| CN110781422B (en) | Page configuration method, page configuration device, computer equipment and storage medium | |
| US20240037142A1 (en) | Systems and methods for filtering of computer vision generated tags using natural language processing | |
| CN109978261A (en) | Determine method, apparatus, readable medium and the electronic equipment of load forecasting model | |
| CN109753884A (en) | A kind of video behavior recognition methods based on key-frame extraction | |
| CN110083386B (en) | Random number generation control method, device, computer equipment and storage medium | |
| CN106708443A (en) | Data reading and writing method and device | |
| CN109635148A (en) | Face picture storage method and device | |
| De Marsico et al. | FACE: Face analysis for commercial entities | |
| CN110222582A (en) | A kind of image processing method and camera | |
| CN115114395A (en) | Content retrieval and model training method and device, electronic equipment and storage medium | |
| CN111476878A (en) | 3D face generation control method and device, computer equipment and storage medium | |
| CN106663191A (en) | System and method for classifying pixels | |
| CN107273435A (en) | Video personnel's fuzzy search parallel method based on MapReduce | |
| CN111061911B (en) | Target detection and tracking method, device and equipment for multi-video monitoring data | |
| CN108875037A (en) | Polymorphic type picture classification sort method and system | |
| US20220230331A1 (en) | Method and electronic device for determining motion saliency and video playback style in video | |
| CN116052251A (en) | Face image optimization method, device, equipment and storage medium | |
| CN110428133A (en) | Personnel's packet control process, device, computer equipment and storage medium | |
| CN109597919A (en) | A kind of data managing method and system merging chart database and intelligent algorithm |
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 |