公开(公告)号：US11636379B2

公开(公告)日：2023-04-25

申请号：US16141886

申请日：2018-09-25

Applicant: Alibaba Group Holding Limited

Inventor： Jun Zhou

IPC: G06N20/00 ,  G06F9/38 ,  G06F9/50

Abstract: A distributed cluster training method and an apparatus thereof are provided. The method includes reading a sample set, the sample set including at least one piece of sample data; using the sample data and current weights to substitute into a target model training function for iterative training to obtain a first gradient before receiving a collection instruction, the collection instruction being issued by a scheduling server when a cluster system environment meets a threshold condition; sending the first gradient to an aggregation server if a collection instruction is received, wherein the aggregation server collects each first gradient and calculates second weights; and receiving the second weights sent by the aggregation server to update current weights. The present disclosure reduces an amount of network communications and an impact on switches, and avoids the use of an entire cluster from being affected.

公开(公告)号：US11544618B2

公开(公告)日：2023-01-03

申请号：US16132264

申请日：2018-09-14

Applicant: Alibaba Group Holding Limited

Inventor： Shenquan Qu ,  Jun Zhou ,  Yongming Ding

IPC: G06F16/00 ,  G06N20/00 ,  G06K9/62 ,  G06F16/9535

Abstract: An automatic multi-threshold feature filtering method and an apparatus thereof are provided. In an iterative process of training a machine learning model, the feature filtering method calculates a feature filtering threshold and feature correlation values of a current round of iteration based on a result of a previous iteration, and performs feature filtering on samples based on the calculated feature filtering threshold and the calculated feature correlation values. The feature filtering apparatus of the present disclosure includes a calculation module and a feature filtering module. The method and apparatus of the present disclosure can automatically generate different feature filtering thresholds at each iteration, which greatly improves an accuracy of a filtering threshold, and can greatly increase the training speed of automatic machine learning and an accuracy of a machine learning model compared with fixed and single thresholds nowadays.

公开(公告)号：US20210011788A1

公开(公告)日：2021-01-14

申请号：US16960302

申请日：2019-01-16

Applicant: Alibaba Group Holding Limited

Inventor： Shaosheng Cao ,  Jun Zhou

IPC: G06F9/54 ,  G06N20/00

Abstract: Embodiments of the present specification disclose a vector-processing method, apparatus, and device for RPC information. The scheme comprises: acquiring an RPC-information sequence consisting of a plurality of RPC-information units of a user; establishing and initializing feature vectors of the RPC-information units; and training the feature vectors according to the RPC-information sequence and the feature vectors, so as to obtain feature vectors with accurate expression.

公开(公告)号：US20200293892A1

公开(公告)日：2020-09-17

申请号：US16886609

申请日：2020-05-28

Applicant: Alibaba Group Holding Limited

Inventor： Jun Zhou

IPC: G06N3/08 ,  G06N3/04

Abstract: A sample is obtained from a test sample set. The sample is input into a plurality of models included in a model set that are to be tested, where the plurality of models include at least one neural network model. A plurality of output results are obtained, including obtaining, from each model of the plurality of models, a respective output result. A test result is determined based on the plurality of output results, where the test result includes at least one of a first test result or a second test result, where the first test result includes a plurality of output result accuracies. In response to determining that the test result does not satisfy a predetermined condition, a new sample is generated based on the sample and a predetermined rule, and the new sample is added to the test sample set.

公开(公告)号：US10740561B1

公开(公告)日：2020-08-11

申请号：US16670533

申请日：2019-10-31

Applicant: Alibaba Group Holding Limited

Inventor： Shaosheng Cao ,  Jun Zhou

IPC: G06F17/27 ,  G06F40/295 ,  G16H15/00 ,  G06K9/00

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for entity prediction. One of the methods includes performing word segmentation on text to be predicted to obtain a plurality of words. For each word of the plurality of words, a determination is made whether the word has a pre-trained word vector. In response to determining that the word has a pre-trained word vector, the pre-trained word vector for the word is obtained. In response to determining that the word does not have a pre-trained word vector, a word vector for the word is determined based on a pre-trained stroke vector. The word vector and the pre-trained stroke vector are trained based on a text sample and a word vector model. An entity associated with the text is predicted by inputting word vectors of the plurality of words into an entity prediction model.

公开(公告)号：US20200226194A1

公开(公告)日：2020-07-16

申请号：US16833021

申请日：2020-03-27

Applicant: Alibaba Group Holding Limited

Inventor： Jun Zhou

IPC: G06F16/955 ,  G06F16/958 ,  G06F16/242 ,  G06F16/00 ,  H04L29/12

Abstract: A server receives a short link application from a requester. The short link application includes a long link uniform resource locator (URL). The server obtains a database identifier based on the long link URL. The server determines whether a database associated with the database identifier is accessible by the server. In response to a determination that the database associated with the database identifier is accessible by the server, the server obtains a short link URL associated with the long link URL from the database, and transmits the short link URL to the requester.

公开(公告)号：US10672058B2

公开(公告)日：2020-06-02

申请号：US16390057

申请日：2019-04-22

Applicant: Alibaba Group Holding Limited

Inventor： Chaochao Chen ,  Jun Zhou

IPC: G06Q30/00 ,  G06Q30/06 ,  G06F21/62 ,  G06Q50/00

Abstract: An item rating and recommendation platform identifies rating data comprising respective ratings of multiple items with respect to multiple users, identifies user-feature data comprising multiple user features contributing to the respective ratings of the multiple items with respect to the multiple users, and receives, from a social network platform via a secret sharing scheme with a trusted initializer, manipulated social network data computed based on social network data and first input data from the trusted initializer. The social network data indicate social relationships between any two of the multiple users. In the secret sharing scheme with the trusted initializer, the social network platform shares with the item rating and recommendation platform the manipulated social network data without disclosing the social network data. The item rating and recommendation platform updates the user-feature data based on the rating data and the manipulated social network data.

公开(公告)号：US20190042763A1

公开(公告)日：2019-02-07

申请号：US16053606

申请日：2018-08-02

Applicant: Alibaba Group Holding Limited

Inventor： Peilin Zhao ,  Jun Zhou ,  Xiaolong Li ,  Longfei Li

IPC: G06F21/60 ,  G06N99/00

Abstract: Techniques for data sharing between a data miner and a data provider are provided. A set of public parameters is downloaded from the data miner. The public parameters are data miner parameters associated with a feature set of training sample data. A set of private parameters in the data provider can be replaced with the set of public parameters. The private parameters are data provider parameters associated with the feature set of training sample data. The private parameters are updated to provide a set of update results. The private parameters are updated based on a model parameter update algorithm associated with the data provider. The update results is uploaded to the data miner.

公开(公告)号：US20180365595A1

公开(公告)日：2018-12-20

申请号：US16119585

申请日：2018-08-31

Applicant: Alibaba Group Holding Limited

Inventor： Jun Zhou

IPC: G06N99/00

Abstract: A training method and a training system for a machine learning system are provided. The method includes allocating training data to a plurality of working machines; dividing training data allocated by each working machine into a plurality of data pieces; obtaining a local weight and a local loss function value calculated by each working machine based on each data piece; aggregating the local weight and the local loss function value calculated by each work machine based on each data piece to obtain a current weight and a current loss function value; performing model abnormality detection using the current weight and/or the current loss function value; inputting a weight and a loss function value of a previous aggregation to the machine learning system for training in response to a result of the model abnormality detection being a first type of abnormality; and modifying the current weight and/or the current loss function value to a current weight and/or a current loss function value within a first threshold in response to the result of the model abnormality detection being a second type of abnormality, and inputting thereof to the machine learning system for training.

公开(公告)号：US20180307773A1

公开(公告)日：2018-10-25

申请号：US16019228

申请日：2018-06-26

Applicant: Alibaba Group Holding Limited

Inventor： Jun Zhou

IPC: G06F17/30

Abstract: A server receives a short link application from a requester. The short link application includes a long link uniform resource locator (URL). The server obtains a database identifier based on the long link URL. The server determines whether a database associated with the database identifier is accessible by the server. In response to a determination that the database associated with the database identifier is accessible by the server, the server obtains a short link URL associated with the long link URL from the database, and transmits the short link URL to the requester.