Feature Transformation Apparatus for Accurate Data Representation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for factorizing data into a sum of features, such as factorial HMM and latent feature models, face challenges with non-uniqueness, leading to inaccurate estimation of feature combinations.

Innovation Solution

A feature transformation apparatus and method that stores feature combinations and transforms them to maintain the sum, using a combination storage part and transformation part to ensure accurate representation of data as a sum of features, even in cases of non-uniqueness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If data is factorized into a sum of feature combinations using factorial HMM or latent feature models, then the data can be represented as a combination of latent features, but the factorization is not unique leading to inaccurate estimation of feature combinations

Engineering Contradiction:

Improveability to represent data as feature combinationVSAvoidaccuracy of feature combination estimation

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies parameter changes by transforming the feature combinations through learned transformation matrices. Instead of directly using the factorized features, the system learns optimal transformation parameters that map the original feature combinations to transformed versions with better identifiability and accuracy, thereby resolving the non-uniqueness problem while preserving the data representation capability

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent implements feedback mechanisms by using the observed data to learn transformation matrices that improve the accuracy of feature combination estimation. The system continuously refines the transformation parameters based on the relationship between observed data and factorized features, creating a feedback loop that enhances measurement precision without sacrificing adaptability

Inventive Principle:

Principle #23Feedback

2Measurement precision

If feature combinations are transformed to improve accuracy, then the estimation precision improves, but the system complexity increases due to additional transformation components

Engineering Contradiction:

Improveaccuracy of feature combination estimationVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex transformation process into manageable components: feature factorization module, transformation matrix learning module, and feature combination estimation module. By dividing the system into independent functional segments, the complexity is distributed and made more tractable while maintaining high estimation precision

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The transformation matrices serve multiple functions: they transform features for accuracy improvement, encode prior knowledge about feature relationships, and enable adaptation to different data types. This multi-functionality reduces the need for separate components, thereby managing system complexity while achieving high precision

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11586703B2Feature transformation apparatus and method, and recording medium

Publication Date: 2023.02.21 NEC CORP

AI summary

A feature transformation apparatus includes at least a combination storage part that stores a combination with respect to a set of features, wherein data is approximately represented as a sum of the combination of the features; and a transformation part that transforms at least the combination so as not to change the sum of the combination of the set of features.