Neural Network Prior Distribution Estimation for Missing Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing matrix decomposition methods require a large amount of observation data to accurately estimate missing values, which is not feasible when limited data is available.

Innovation Solution

A learning method using a neural network to estimate parameters of prior distributions for updated data sets with missing values, followed by parameter updates to enhance missing value estimation accuracy, allowing for accurate estimation with a smaller number of observation data points.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If matrix decomposition is used to estimate missing values, then missing value estimation can be performed, but a large amount of observation data is required which is not available in many cases

Engineering Contradiction:

Improvemissing value estimation accuracyVSAvoidamount of observation data

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent changes the approach from traditional matrix decomposition to a neural network-based method that learns parameters of prior distributions. This parameter transformation enables the system to achieve accurate missing value estimation with limited observation data by capturing complex non-linear relationships and patterns that traditional linear methods cannot handle effectively.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces the traditional mechanical/mathematical system of matrix decomposition with a neural network-based computational system. This substitution allows the system to process limited data more effectively by using learned representations and prior distributions, achieving superior estimation accuracy when data is scarce.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Ease of manufacture

If traditional matrix decomposition methods are used, then the method is simple and well-established, but estimation accuracy deteriorates when observation data is limited

Engineering Contradiction:

Improvemethod simplicityVSAvoidmissing value estimation accuracy

Core Design Contradiction:

Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent introduces an intermediary layer of neural network parameters that mediates between the limited observation data and the missing value estimation. These parameters act as a bridge, capturing complex patterns from the available data and using prior distributions to infer missing values accurately, even when the observation data is insufficient for traditional methods.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20240169204A1Learning method, estimation method, learning apparatus, estimation apparatus, and program

Publication Date: 2024.05.23 NIPPON TELEGRAPH & TELEPHONE CORP

AI summary

A learning method executed by a computer including a memory and a processor, that includes: inputting a learning data set including a plurality of pieces of observation data; estimating, by a neural network, parameters of prior distributions of a plurality of pieces of data in a case where the post-missing observation data is expressed by a product of the plurality of pieces of data, using the post-missing observation data in which some values included in the observation data are set as missing values; updating the plurality of pieces of data using the parameters of the prior distributions such that the product of the plurality of pieces of data matches the post-missing observation data; estimating a missing value of the post-missing observation data from the plurality of pieces of updated data; and updating model parameters including parameters of the neural network to increase estimation accuracy of the missing value.