DNN Feature Vectors for Arbitrary-Range Class Explanations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep neural network (DNN) technologies lack the ability to generate descriptive information for arbitrary ranges designated by a user across the entire input data, limiting their interpretability in class discrimination processes.
Innovation Solution
A method and device for generating descriptive information using a convolutional neural network with residual blocks, involving steps to input and process input data, generate feature vectors, associate positions with feature maps, and calculate similarities between designated data ranges and classes, enabling detailed class classification analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If similarity calculation is performed only when small regions similar to prototype features are found, then calculation efficiency is improved, but the ability to generate descriptive information for arbitrary user-designated ranges is lost
Solution Approach 1:
The patent pre-calculates and stores feature vectors for all possible regions in the input data during a preprocessing stage. This preliminary action enables the system to quickly retrieve and compare feature vectors for any user-designated range without performing complex calculations at query time, thus maintaining both efficiency and versatility
Solution Approach 2:
The system dynamically adapts its behavior based on the user's query requirements. When a user designates an arbitrary range, the system flexibly retrieves pre-computed feature vectors corresponding to that specific range and performs similarity calculations only for the requested region, rather than following a fixed calculation pattern
2Loss of information
If feature vectors are extracted from the entire feature map, then comprehensive class discrimination information is obtained, but the ability to analyze specific arbitrary ranges is reduced
Solution Approach 1:
The patent segments the feature map into multiple regions and extracts feature vectors for each segment independently. This segmentation allows the system to maintain comprehensive information from the entire feature map while also enabling easy extraction and analysis of specific arbitrary ranges by simply selecting the relevant segmented regions
Solution Approach 2:
The system organizes feature vectors in a multi-dimensional structure that includes both spatial position information and feature data. This dimensional organization allows simultaneous access to comprehensive information across the entire feature map and efficient querying of specific ranges by filtering on spatial coordinates
Data Source
AI summary
A method of generating descriptive information regarding class classification includes a step of inputting input data to a machine learning model again and acquiring a set of M×N first feature vectors corresponding to a size of the first feature maps from L first feature maps that are outputs of a specific layer, a step of generating, for each class group, a known feature vector including a set of the first feature vectors, a step of receiving a designation of a range in discrimination target data, a step of inputting the discrimination target data to the machine learning model and acquiring a set of M×N second feature vectors corresponding to the size of the second feature map from a second feature map that is an output of the specific layer, and a step of calculating a similarity between the designated range in the discrimination target data and at least one of the classes using the set of second feature vectors and the known feature vector group of at least one of the classes.