Sparse Feature Vector Generation for Classification

Solution Overview

Problem

Existing classification systems face challenges with high computational complexity and memory requirements due to dense feature vectors, which can lead to reduced classification accuracy and increased latency.

Innovation Solution

The implementation of enforced sparsity in feature extraction, where only the top K percentage of feature values are retained, producing a sparse feature vector with fewer non-zero elements, thereby reducing memory usage and improving classifier performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If dense feature vectors are used for classification, then classification accuracy can be maintained, but memory consumption and computational complexity increase significantly

Engineering Contradiction:

Improveclassification accuracyVSAvoidmemory consumption

Core Design Contradiction:

ReliabilityVSQuantity of substance

Solution Approach 1:

The patent extracts only the top K percentage of most significant feature values from the dense feature vector, discarding the remaining less significant values. This extraction creates a sparse feature vector that retains the most important classification information while eliminating redundant data, thereby reducing memory consumption without significantly compromising classification accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the sparsity parameter of the feature vector by setting a threshold percentage (e.g., top 10%, top 5%) of feature values to retain. By adjusting this sparsity parameter, the system optimizes the balance between maintaining sufficient classification information and reducing memory usage, transforming dense vectors into sparse vectors with controlled non-zero elements.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If dense feature vectors are used for classification, then comprehensive feature information is preserved, but computational complexity and processing time increase

Engineering Contradiction:

Improveclassification accuracyVSAvoidcomputational complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts only the top K percentage of most significant feature values from the dense feature vector, discarding the remaining less significant values. This extraction creates a sparse feature vector that retains the most important classification information while eliminating redundant data, thereby reducing memory consumption without significantly compromising classification accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the sparsity parameter of the feature vector by setting a threshold percentage (e.g., top 10%, top 5%) of feature values to retain. By adjusting this sparsity parameter, the system optimizes the balance between maintaining sufficient classification information and reducing memory usage, transforming dense vectors into sparse vectors with controlled non-zero elements.

Inventive Principle:

Principle #35Parameter changes

3Reliability

If more feature values are retained in the feature vector, then classification accuracy improves, but memory usage and processing time increase

Engineering Contradiction:

Improveclassification accuracyVSAvoidprocessing speed

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent extracts only the top K percentage of most significant feature values from the dense feature vector, discarding the remaining less significant values. This extraction creates a sparse feature vector that retains the most important classification information while eliminating redundant data, thereby reducing memory consumption without significantly compromising classification accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the sparsity parameter of the feature vector by setting a threshold percentage (e.g., top 10%, top 5%) of feature values to retain. By adjusting this sparsity parameter, the system optimizes the balance between maintaining sufficient classification information and reducing memory usage, transforming dense vectors into sparse vectors with controlled non-zero elements.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS11423323B2Generating a sparse feature vector for classification

Publication Date: 2022.08.23 QUALCOMM INC

AI summary

An apparatus for classifying an input includes a classifier and a feature extractor. The feature extractor is configured to generate a feature vector based on the input. The feature vector is also configured to set a number of elements of the feature vector to zero to produce a sparse feature vector. The sparse feature vector has the same dimensions as the feature vector generated by the feature extractor. However, the sparse feature vector includes fewer non-zero elements than the feature vector generated by the feature extractor. The feature vector is further configured to forward the sparse feature vector to the classifier to classify the input.