WO2017157183A1 - 一种自动多阀值特征过滤方法及装置 - Google Patents

一种自动多阀值特征过滤方法及装置 Download PDF

Info

Publication number
WO2017157183A1
WO2017157183A1 PCT/CN2017/075517 CN2017075517W WO2017157183A1 WO 2017157183 A1 WO2017157183 A1 WO 2017157183A1 CN 2017075517 W CN2017075517 W CN 2017075517W WO 2017157183 A1 WO2017157183 A1 WO 2017157183A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature
threshold
filtering
dimension
gradient
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.)
Ceased
Application number
PCT/CN2017/075517
Other languages
English (en)
French (fr)
Chinese (zh)
Inventor
瞿神全
周俊
崔卿
丁永明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Alibaba Group Holding Ltd
Original Assignee
Alibaba Group Holding Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Alibaba Group Holding Ltd filed Critical Alibaba Group Holding Ltd
Priority to JP2018548836A priority Critical patent/JP6890607B2/ja
Publication of WO2017157183A1 publication Critical patent/WO2017157183A1/zh
Priority to US16/132,264 priority patent/US11544618B2/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06F18/211—Selection of the most significant subset of features
    • G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation

Definitions

  • the invention belongs to the field of artificial intelligence technology, and in particular relates to an automatic multi-threshold feature filtering method and device.
  • the ultra-large-scale machine learning algorithm is the basic technical support for the current Internet companies to achieve search query ranking, Internet advertising click rate prediction, product personalized recommendation, speech recognition, intelligent question and answer system.
  • the ever-expanding data scale has brought about great theoretical and engineering challenges for large-scale data processing while improving the application of algorithms. Efficient data processing has become the core technology of Internet big data applications.
  • Internet data is usually very sparse, so after training the machine learning model using Internet data, a sparse model is obtained, and the sparse model facilitates subsequent storage and prediction.
  • Applying efficient feature filtering algorithm in extremely sparse Internet data training can effectively remove unrelated features and redundant features, improve the generalization performance and operational efficiency of learning algorithms, and is of great help for training machine learning models.
  • a sample feature ⁇ feature_1, feature_2,...,feature_n ⁇ ;
  • a value x is calculated according to a certain dimension gradient value gi and other model parameters, and the magnitudes of x and r are compared. If x is smaller than r, the dimension feature is filtered out.
  • the threshold used for filtering in the existing feature filtering method needs to be manually specified, so the filtering effect is strongly dependent. Human experience, the filtering effect is unstable, because the filtering threshold cannot be automatically adjusted according to the sample data, which leads to the prediction accuracy of the model obtained by the training; and the threshold is only one and fixed, and the threshold cannot be dynamically adjusted according to the training situation, so it cannot be very Filter all features well. Since the threshold value is not well determined, if the threshold is not well selected, the machine learning model obtained by the training is unreliable and the prediction accuracy is lowered.
  • the object of the present invention is to provide an automatic multi-threshold feature filtering method and device, so as to solve the prior art feature filtering method, which can only use a fixed threshold for feature filtering, resulting in low training efficiency and insufficient machine learning model for training. The exact problem.
  • An automatic multi-threshold feature filtering method for feature filtering of sample data during training of a machine learning model comprising:
  • the feature filtering is performed on the sample according to the calculated feature filtering threshold and the feature correlation value.
  • the characteristic filter threshold is a ratio of a maximum gradient in the gradient obtained by each dimension feature in the sample data to the number of sample data in the previous iteration.
  • the calculating the feature correlation value of the current iteration according to the result of the previous round of iteration including:
  • Corresponding feature correlation values are calculated according to the gradient of each dimension feature.
  • the feature correlation value of each dimension feature is a linear function of the gradient of each dimension.
  • performing feature filtering on the sample according to the calculated feature filtering threshold and the feature correlation value including:
  • the dimension feature is filtered out and does not participate in subsequent iterative calculations; otherwise, the dimension feature is retained and continues to participate in subsequent iterative calculations.
  • the invention also provides an automatic multi-threshold feature filtering device for feature filtering of sample data during training of a machine learning model, the feature filtering device comprising:
  • a calculation module configured to calculate a feature filtering threshold and a feature correlation value of the current iteration according to a result of the previous iteration
  • the feature filtering module is configured to perform feature filtering on the sample according to the calculated feature filtering threshold and the feature correlation value.
  • the characteristic filter threshold is a gradient obtained by each dimension feature in the sample data during the previous iteration The ratio of the maximum gradient to the number of sample data.
  • calculation module calculates the feature correlation value of the current iteration according to the result of the previous iteration.
  • Corresponding feature correlation values are calculated according to the gradient of each dimension feature.
  • the feature correlation value of each dimension feature is a linear function of the gradient of each dimension.
  • the feature filtering module performs feature filtering on the sample according to the calculated feature filtering threshold and the feature correlation value, and performs the following steps:
  • the dimension feature is filtered out and does not participate in subsequent iterative calculations; otherwise, the dimension feature is retained and continues to participate in subsequent iterative calculations.
  • the invention provides an automatic multi-threshold feature filtering method and device, which breaks through the existing method of artificially setting a single threshold for feature filtering, and can automatically calculate multiple thresholds according to the iterative result of each batch of sample data to filter features. Greatly improved the training speed and the accuracy of the machine learning model obtained by training.
  • FIG. 1 is a flow chart of an automatic multi-threshold feature filtering method according to the present invention.
  • FIG. 2 is a schematic structural view of an automatic multi-threshold feature filtering device of the present invention.
  • a machine learning model is trained using a large amount of original sample data.
  • the sample data has multi-dimensional features, such as price, product category, etc. These features have different effects on the improvement. Some features may not improve the effect. You can filter out this feature, and the effective features will remain.
  • the remaining features are finally weighted by training, and these weights are the model parameters corresponding to the obtained machine learning model.
  • the general idea of the present invention is to calculate feature filter values according to current model parameters during each iteration of machine learning model training, and use the calculated feature filter values to perform feature filtering.
  • the automatic multi-threshold feature filtering method in this embodiment includes:
  • Step S1 Calculate the feature filtering threshold and the feature correlation value of the current iteration according to the result of the previous round of iteration.
  • This embodiment takes a typical machine learning process as an example, assuming that the estimated function of the machine learning model is:
  • ⁇ is the model parameter and x is the sample feature, both of which are vectors, and x i is the i-dimensional feature.
  • the loss function J( ⁇ ) is also defined in machine learning to evaluate whether ⁇ is better, and ⁇ is adjusted such that J( ⁇ ) takes a minimum value. In order to achieve this, it is necessary to iterate according to the least squares method or the gradient descent method until the final convergence obtains a value of ⁇ such that J( ⁇ ) is the smallest.
  • the gradient descent method is taken as an example, and the formula for calculating the k-th gradient g k is as follows:
  • the training process and the gradient descent method for the machine learning model are not described here.
  • the present embodiment uses the iterative result in the above process to calculate the feature filtering threshold.
  • the specific calculation method is as follows:
  • l is the number of samples
  • g (k-1)i is the gradient value corresponding to the i-th dimension of the k-1th round.
  • the filter is calculated according to the number of threshold r k l samples of the original sample data and the gradients g k
  • the calculation may be implemented using a variety of algorithms, not dependent on a particular algorithm.
  • the calculation may be performed according to the gradient g k and the dimension of the sample feature, or may be calculated according to the gradient g k and the saliency parameter of the sample feature, which will not be repeated here.
  • the feature correlation value s ki corresponding to the i-th dimension of the k-th wheel needs to be calculated, and the calculation formula is as follows:
  • ⁇ is a fixed constant. It can be seen that the feature filtering threshold r k of the embodiment is calculated according to the gradient g (k-1)i of the previous iteration, and the feature correlation value is calculated according to the gradient g ki of the current round, and the feature correlation value of each dimension feature is each dimension.
  • the linear function of the gradient is a fixed constant.
  • the calculation parameters of the feature filtering threshold and the feature correlation value need to be unified, that is, the feature filtering threshold is calculated according to the gradient, and the feature correlation value is also calculated according to the gradient, but the specific calculation formula is different according to the training model. It can be designed differently, even if it is training the same model can be designed differently.
  • the feature correlation value of this embodiment is different in that different feature correlation values are calculated for each iteration.
  • the invention is not limited to the specific method of calculating the characteristic filtering threshold and the feature correlation value.
  • Step S2 Perform feature filtering on the sample according to the calculated feature filtering threshold and the feature correlation value.
  • step S1 performs feature filtering on the sample by comparing the size between the feature filtering threshold g ki and the filtering threshold s ki . specifically:
  • the i-th dimension feature is filtered out and does not participate in subsequent calculations; otherwise, the i-th dimension feature is retained and continues to participate in subsequent calculations.
  • the original sample data will be filtered out of the features of the partial dimensions, and the filtered data will be imported into the system as the new sample data for the next iteration until the iteration termination condition is reached.
  • the gradient g ki is calculated based on the sample data, the loss function, and the model parameters, The sample data and the model parameters in the round iteration are different, so the correlation filter threshold value calculated in each iteration and the correlation value s ki of the per-dimensional feature of the sample data are different.
  • the termination condition of the iterative calculation according to a specific algorithm, for example, no new features are filtered out after the last iteration, or the number of iterations exceeds the set maximum value, which will not be described here.
  • the features that are retained after the iteration is completed finally get different model parameters through training, and the machine learning model is obtained according to these model parameters.
  • the present embodiment simultaneously provides an automatic multi-threshold feature filtering device for feature filtering of sample data during machine learning model training, the device comprising:
  • a calculation module configured to calculate a feature filtering threshold and a feature correlation value of the current iteration according to a result of the previous iteration
  • the feature filtering module is configured to perform feature filtering on the sample according to the calculated feature filtering threshold and the feature correlation value.
  • the feature filter threshold is a ratio of the maximum gradient in the gradient obtained by the dimension features in the sample data to the number of sample data in the previous iteration.
  • the calculation module calculates the feature filtering threshold and the feature correlation value of the current iteration according to the result of the previous iteration, wherein the calculation module performs the following steps when calculating the feature correlation value of the current iteration according to the result of the previous iteration:
  • Corresponding feature correlation values are calculated according to the gradient of each dimension feature.
  • the feature correlation value s ki corresponding to the i-th feature of the kth round of the embodiment is calculated according to the formula 4, and the feature correlation value of each dimension feature is a linear function of the gradient of each dimension.
  • the feature filtering module performs feature filtering on the sample according to the calculated feature filtering threshold and the feature correlation value, and performs the following steps:
  • the dimension feature is filtered out and does not participate in subsequent iterative calculations; otherwise, the dimension feature is retained and continues to participate in subsequent iterative calculations.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Business, Economics & Management (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Biology (AREA)
  • Accounting & Taxation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Image Analysis (AREA)
PCT/CN2017/075517 2016-03-17 2017-03-03 一种自动多阀值特征过滤方法及装置 Ceased WO2017157183A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2018548836A JP6890607B2 (ja) 2016-03-17 2017-03-03 自動多閾値特徴フィルタリング方法及び装置
US16/132,264 US11544618B2 (en) 2016-03-17 2018-09-14 Automatic multi-threshold feature filtering method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610154296.1 2016-03-17
CN201610154296.1A CN107203891A (zh) 2016-03-17 2016-03-17 一种自动多阀值特征过滤方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/132,264 Continuation US11544618B2 (en) 2016-03-17 2018-09-14 Automatic multi-threshold feature filtering method and apparatus

Publications (1)

Publication Number Publication Date
WO2017157183A1 true WO2017157183A1 (zh) 2017-09-21

Family

ID=59850590

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/075517 Ceased WO2017157183A1 (zh) 2016-03-17 2017-03-03 一种自动多阀值特征过滤方法及装置

Country Status (5)

Country Link
US (1) US11544618B2 (https=)
JP (1) JP6890607B2 (https=)
CN (1) CN107203891A (https=)
TW (1) TWI794157B (https=)
WO (1) WO2017157183A1 (https=)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3926547A1 (en) 2020-06-18 2021-12-22 Fujitsu Limited Program, learning method, and information processing apparatus
EP3929823A1 (en) 2020-06-25 2021-12-29 Fujitsu Limited Program, learning method, and information processing apparatus
TWI790769B (zh) * 2021-10-07 2023-01-21 宏碁股份有限公司 電子郵件備份方法與電子郵件管理系統

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10609475B2 (en) 2014-12-05 2020-03-31 Stages Llc Active noise control and customized audio system
US10331802B2 (en) 2016-02-29 2019-06-25 Oracle International Corporation System for detecting and characterizing seasons
US10699211B2 (en) 2016-02-29 2020-06-30 Oracle International Corporation Supervised method for classifying seasonal patterns
US10867421B2 (en) 2016-02-29 2020-12-15 Oracle International Corporation Seasonal aware method for forecasting and capacity planning
US10885461B2 (en) 2016-02-29 2021-01-05 Oracle International Corporation Unsupervised method for classifying seasonal patterns
US10198339B2 (en) 2016-05-16 2019-02-05 Oracle International Corporation Correlation-based analytic for time-series data
US11082439B2 (en) 2016-08-04 2021-08-03 Oracle International Corporation Unsupervised method for baselining and anomaly detection in time-series data for enterprise systems
US10635563B2 (en) 2016-08-04 2020-04-28 Oracle International Corporation Unsupervised method for baselining and anomaly detection in time-series data for enterprise systems
US10945080B2 (en) 2016-11-18 2021-03-09 Stages Llc Audio analysis and processing system
US10915830B2 (en) 2017-02-24 2021-02-09 Oracle International Corporation Multiscale method for predictive alerting
US10949436B2 (en) 2017-02-24 2021-03-16 Oracle International Corporation Optimization for scalable analytics using time series models
US10817803B2 (en) * 2017-06-02 2020-10-27 Oracle International Corporation Data driven methods and systems for what if analysis
US10963346B2 (en) 2018-06-05 2021-03-30 Oracle International Corporation Scalable methods and systems for approximating statistical distributions
US10997517B2 (en) 2018-06-05 2021-05-04 Oracle International Corporation Methods and systems for aggregating distribution approximations
US12001926B2 (en) 2018-10-23 2024-06-04 Oracle International Corporation Systems and methods for detecting long term seasons
US11138090B2 (en) 2018-10-23 2021-10-05 Oracle International Corporation Systems and methods for forecasting time series with variable seasonality
US11526778B2 (en) * 2018-12-19 2022-12-13 T-Mobile Usa, Inc. Future user device preference prediction based on telecom data
US10855548B2 (en) 2019-02-15 2020-12-01 Oracle International Corporation Systems and methods for automatically detecting, summarizing, and responding to anomalies
US11533326B2 (en) 2019-05-01 2022-12-20 Oracle International Corporation Systems and methods for multivariate anomaly detection in software monitoring
US11537940B2 (en) 2019-05-13 2022-12-27 Oracle International Corporation Systems and methods for unsupervised anomaly detection using non-parametric tolerance intervals over a sliding window of t-digests
US11887015B2 (en) 2019-09-13 2024-01-30 Oracle International Corporation Automatically-generated labels for time series data and numerical lists to use in analytic and machine learning systems
CN110825966B (zh) * 2019-10-31 2022-03-04 广州市百果园信息技术有限公司 一种信息推荐的方法、装置、推荐服务器和存储介质
US12288160B2 (en) 2021-06-16 2025-04-29 International Business Machines Corporation Transfer learning with basis scaling and pruning

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1851729A (zh) * 2006-06-01 2006-10-25 北京中星微电子有限公司 应用于模式识别的基于AdaBoost的特征抽取方法
CN102629305A (zh) * 2012-03-06 2012-08-08 上海大学 一种面向snp数据的特征选择方法
CN103955714A (zh) * 2014-04-09 2014-07-30 中国科学院信息工程研究所 基于水军检测模型构建方法和系统及水军检测方法
CN104616031A (zh) * 2015-01-22 2015-05-13 哈尔滨工业大学深圳研究生院 迁移学习方法及装置

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050281457A1 (en) * 2004-06-02 2005-12-22 Murat Dundar System and method for elimination of irrelevant and redundant features to improve cad performance
JP5207870B2 (ja) * 2008-08-05 2013-06-12 日立コンピュータ機器株式会社 次元削減方法、パターン認識用辞書生成装置、及びパターン認識装置
KR100974900B1 (ko) * 2008-11-04 2010-08-09 한국전자통신연구원 동적 임계값을 이용한 마커 인식 장치 및 방법
JP2014160457A (ja) * 2013-02-20 2014-09-04 Nec Corp 対話的変数選択装置、対話的変数選択方法および対話的変数選択プログラム
US20150206069A1 (en) * 2014-01-17 2015-07-23 Matthew BEERS Machine learning-based patent quality metric
US10540606B2 (en) * 2014-06-30 2020-01-21 Amazon Technologies, Inc. Consistent filtering of machine learning data
CN104504373A (zh) * 2014-12-18 2015-04-08 电子科技大学 一种用于fmri数据的特征选择方法
US9275347B1 (en) * 2015-10-09 2016-03-01 AlpacaDB, Inc. Online content classifier which updates a classification score based on a count of labeled data classified by machine deep learning
CN105389480B (zh) * 2015-12-14 2018-03-02 深圳大学 多类不平衡基因组学数据迭代集成特征选择方法及系统
WO2017120579A1 (en) * 2016-01-10 2017-07-13 Presenso, Ltd. System and method for validating unsupervised machine learning models

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1851729A (zh) * 2006-06-01 2006-10-25 北京中星微电子有限公司 应用于模式识别的基于AdaBoost的特征抽取方法
CN102629305A (zh) * 2012-03-06 2012-08-08 上海大学 一种面向snp数据的特征选择方法
CN103955714A (zh) * 2014-04-09 2014-07-30 中国科学院信息工程研究所 基于水军检测模型构建方法和系统及水军检测方法
CN104616031A (zh) * 2015-01-22 2015-05-13 哈尔滨工业大学深圳研究生院 迁移学习方法及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YANG, FEIHU: "Research on Feature Selection Approach and its Application in Network Traffic Identification", ELECTRONIC TECHNOLOGY & INFORMATION SCIENCE , CHINA MASTER'S THESES FULL-TEXT DATABASE, 15 July 2012 (2012-07-15), pages 15 - 22, ISSN: 1674-0246 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3926547A1 (en) 2020-06-18 2021-12-22 Fujitsu Limited Program, learning method, and information processing apparatus
EP3929823A1 (en) 2020-06-25 2021-12-29 Fujitsu Limited Program, learning method, and information processing apparatus
TWI790769B (zh) * 2021-10-07 2023-01-21 宏碁股份有限公司 電子郵件備份方法與電子郵件管理系統

Also Published As

Publication number Publication date
CN107203891A (zh) 2017-09-26
TWI794157B (zh) 2023-03-01
US11544618B2 (en) 2023-01-03
JP2019513265A (ja) 2019-05-23
TW201734840A (zh) 2017-10-01
JP6890607B2 (ja) 2021-06-18
US20190042982A1 (en) 2019-02-07

Similar Documents

Publication Publication Date Title
TWI794157B (zh) 自動多閾值特徵過濾方法及裝置
CN110084221B (zh) 一种基于深度学习的带中继监督的序列化人脸关键点检测方法
CN111091199B (zh) 一种基于差分隐私的联邦学习方法、装置及存储介质
CN114596497B (zh) 目标检测模型的训练方法、目标检测方法、装置及设备
CN109360581B (zh) 基于神经网络的语音增强方法、可读存储介质及终端设备
CN106022392B (zh) 一种深度神经网络样本自动取舍的训练方法
CN105678253B (zh) 半监督人脸年龄估计装置及半监督人脸年龄估计方法
CN107358293A (zh) 一种神经网络训练方法及装置
CN107506350B (zh) 一种识别信息的方法和设备
CN110291540A (zh) 批再归一化层
CN107564513A (zh) 语音识别方法及装置
CN108596204B (zh) 一种基于改进型scdae的半监督调制方式分类模型的方法
CN117271808B (zh) 一种基于信息瓶颈去噪的多媒体推荐方法
CN110069781B (zh) 一种实体标签的识别方法及相关设备
CN106897265B (zh) 词向量训练方法及装置
CN114399808B (zh) 一种人脸年龄估计方法、系统、电子设备及存储介质
CN110097060A (zh) 一种面向树干图像的开集识别方法
CN111985279B (zh) 多元属性预测模型的训练方法、装置和存储介质
CN108985348A (zh) 基于卷积神经网络的书法风格识别方法
WO2021035980A1 (zh) 人脸识别模型的训练方法、装置、设备及可读存储介质
CN114385831B (zh) 一种基于特征提取的知识图谱关系预测方法
CN105740354A (zh) 自适应潜在狄利克雷模型选择的方法及装置
CN113205856B (zh) 一种微生物宏基因组分箱方法及系统
CN112348102B (zh) 一种基于查询的自底向上视频定位方法和系统
CN113139464B (zh) 一种电网故障检测方法

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2018548836

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17765725

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 17765725

Country of ref document: EP

Kind code of ref document: A1