GAP Layer Directional Contribution Analysis in Class Discriminant Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing class discriminant models using global average pooling (GAP) layers fail to quantify the contribution of feature vectors to the direction of the GAP layer vector, which significantly influences the class discriminant result.
Innovation Solution
A method and computer program to calculate the degree of contribution of each partial region in the immediately preceding layer to the direction of the GAP layer vector by extracting feature vectors and using cosine similarity to determine the influence on the class discriminant result.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cosine similarity between feature vectors and GAP layer vector is calculated to determine contribution degree, then the contribution of each partial region to the direction of GAP layer vector can be quantified, but the existing method only measures contribution to magnitude rather than direction
Solution Approach 1:
The patent changes the measurement parameter from magnitude-based contribution degree to direction-based contribution degree. Specifically, it uses cosine similarity to measure the angular relationship between feature vectors and the GAP layer vector, thereby capturing directional contribution information that was previously lost. This parameter change enables precise measurement of how each partial region contributes to the direction of the GAP layer vector, which is critical for class discriminant problems.
2Measurement precision
If the degree of contribution to the magnitude of the GAP layer vector is obtained, then the contribution of feature vectors can be quantified, but the direction information reflected in class discriminant results cannot be captured
Solution Approach 1:
The patent transitions from measuring contribution to magnitude to measuring contribution to direction by utilizing cosine similarity. This parameter change allows the system to capture how each partial region's feature vector aligns with the GAP layer vector's direction, which directly impacts class discriminant results. The cosine similarity metric specifically measures angular relationship, thereby preserving directional information that magnitude-based measurements cannot capture.
Data Source
AI summary
A method of the present disclosure includes (a) a step for inputting input data to a class discriminant model to obtain an operation result of the class discriminant model, (b) a step for extracting a plurality of feature vectors in a plurality of partial regions constituting an immediately preceding layer disposed immediately before a GAP layer, and a GAP layer vector being output of the GAP layer, and (c) a step for calculating a degree of contribution of each of the plurality of partial regions related to a direction of the GAP layer vector using the plurality of feature vectors and the GAP layer vector.