Method, device and computer storage medium for data analysis

Patent metadata

Field	Value
Publication number	US-12469268-B2
Application number	US-202017433941-A
Country	US
Kind code	B2
Filing date	Jan 21, 2020
Priority date	Feb 25, 2019
Publication date	Nov 11, 2025
Grant date	Nov 11, 2025

Abstract

Embodiments of the present disclosure relate to a method, a device and a computer storage medium for data analysis. The method comprises: obtaining a prediction model, a processing layer of the prediction model comprising a plurality of processing units, parameters of each of the a plurality of processing units satisfying an objective parameter distribution, an output of the prediction model being determined based on a plurality of groups of parameters determined from the parameter distribution; and applying model input data to the prediction model, so as to obtain a prediction for the model input data. In this way, a more accurate prediction result may be obtained.

First claim

We claim:

1 . A method for data analysis, comprising:
obtaining a prediction model, a processing layer of the prediction model comprising a plurality of processing units, parameters of each of the plurality of processing units satisfying an objective parameter distribution, an output of the prediction model being determined based on a plurality of groups of parameters determined from the objective parameter distribution; obtaining the objective parameter distribution for the processing layer of the prediction model; initializing the parameters of the processing layer of the prediction model based on an initial parameter distribution; determining a group of input data for the processing layer based on a group of training samples; determining a plurality of groups of output data of the processing layer based on the group of input data and the initialized parameter; adjusting the initial parameter distribution based on the plurality of groups of output data, so as to obtain the objective parameter distribution that is close to a ground-truth parameter distribution, wherein the objective parameter distribution causes predictions for the group of training samples by the prediction model to be close to annotation data of the group of training samples, wherein the objective parameter distribution indicates a value range of the parameters, and wherein the annotation data comprises ground-truth values corresponding to the group of training samples; applying model input data to the prediction model, so as to obtain a prediction for the model input data; determining an uncertainty measurement of the prediction at least based on the objective parameter distribution, wherein the uncertainty measurement comprises a SOFTMAX confidence score and a Bayesian Active Learning by Disagreement (BALD) measurement; and generating a comparison between the SOFTMAX confidence score and the BALD measurement.

2 . The method according to claim 1 , wherein obtaining the prediction comprises:
obtaining a plurality of groups of parameter sets associated with the plurality of processing units of the processing layer based on the parameter distribution; determining the plurality of groups of output data of the processing layer corresponding to the plurality of groups of parameter sets; and determining the prediction of the prediction model based on the plurality of groups of output data.

3 . The method according to claim 1 , wherein the prediction model is based on a convolutional neural network, and the processing layer comprises at least one of a convolutional layer and a fully connection layer of the convolutional neural network.

4 . The method according to claim 1 , wherein the model input data is non-time series data.

5 . The method according to claim 1 , further comprising:
providing an indication about the uncertainty measurement.

6 . The method according to claim 1 , further comprising:
in response to the uncertainty measurement being within a predetermined abnormal range, causing the model input data to be annotated; and training the prediction model with the annotated model input data.

7 . The method according to claim 1 , wherein the uncertainty measurement comprises an information entropy.

8 . A device for data analysis, comprising:
at least one processing unit; and at least one memory, coupled to the at least one processing unit and storing instructions executed by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the device to perform acts comprising: obtaining a prediction model, a processing layer of the prediction model comprising a plurality of processing units, parameters of each of the plurality of processing units satisfying an objective parameter distribution, an output of the prediction model being determined based on a plurality of groups of parameters determined from the objective parameter distribution; obtaining the objective parameter distribution for the processing layer of the prediction model; initializing the parameters of the processing layer of the prediction model based on an initial parameter distribution; determining a group of input data for the processing layer based on a group of training samples; determining a plurality of groups of output data of the processing layer based on the group of input data and the initialized parameter; adjusting the initial parameter distribution based on the plurality of groups of output data, so as to obtain the objective parameter distribution that is close to a ground-truth parameter distribution, wherein the objective parameter distribution causes predictions for the group of training samples by the prediction model to be close to annotation data of the group of training samples, wherein the objective parameter distribution indicates a value range of the parameters, and wherein the annotation data comprises ground-truth values corresponding to the group of training samples; applying model input data to the prediction model, so as to obtain a prediction for the model input data; determining an uncertainty measurement of the prediction at least based on the objective parameter distribution, wherein the uncertainty measurement comprises a SOFTMAX confidence score and a Bayesian Active Learning by Disagreement (BALD) measurement; and generating a comparison between the SOFTMAX confidence score and the BALD measurement.

9 . The device according to claim 8 , wherein obtaining the prediction comprises:
obtaining a plurality of groups of parameter sets associated with the plurality of processing units of the processing layer based on the parameter distribution; determining the plurality of groups of output data of the processing layer corresponding to the plurality of groups of parameter sets; and determining the prediction of the prediction model based on the plurality of groups of output data.

10 . The device according to claim 8 , wherein the prediction model is based on a convolutional neural network, and the processing layer comprises at least one of a convolutional layer and a fully connection layer of the convolutional neural network.

11 . The device according to claim 8 , wherein the model input data is non-time series data.

12 . The device according to claim 8 , the acts further comprising:
providing an indication about the uncertainty measurement.

13 . The device according to claim 8 , the acts further comprising:
in response to the uncertainty measurement being within a predetermined abnormal range, causing the model input data to be annotated; and training the prediction model with the annotated model input data.

14 . A computer-readable storage medium having computer-readable program instructions stored thereon for performing a method according to claim 1 .

Assignees

• Nec Corp

Inventors

• Zhang Ni
• Chen Xiaoyi

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V10/84

  using probabilistic graphical models from image or video features, e.g. Markov models or Bayesian networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/091

  Active learning · CPC title