Learning Device Feature Vector Similarity Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods for improving robustness against adversarial examples face accuracy deterioration due to averaging feature vectors from multiple models, which can result in feature values being too close for different data types.
Innovation Solution
A learning device and method that calculate multiple feature vectors using various models, then learn parameters based on an evaluation function that minimizes similarity between the average feature vector and class representative vectors, enhancing robustness and accuracy by increasing the distance between feature vectors from different classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If average values of feature vectors from multiple models are calculated to improve robustness against adversarial examples, then robustness is improved, but calculation accuracy deteriorates because feature values become too close for different data types
Solution Approach 1:
The patent changes the parameter of feature vector representation by introducing class-specific representative vectors and adjusting the similarity calculation to account for class differences. This allows the system to maintain robustness while improving accuracy by adapting the feature space to reflect class-specific characteristics.
Solution Approach 2:
The patent applies local quality by introducing class-specific representative vectors that capture the characteristics of each class individually. Instead of using a single average feature vector, the system creates localized representations for each class, allowing for more precise discrimination between different data types while maintaining robustness.
2Reliability
If feature vectors from multiple models are averaged to reduce influence of adversarial examples, then robustness is improved, but the distance between feature vectors of different classes decreases
Solution Approach 1:
The patent changes the parameter of feature representation by introducing class-specific representative vectors. This allows the system to maintain appropriate distances between different classes while still benefiting from the robustness of multi-model averaging, as the class-specific vectors prevent class collapse.
Solution Approach 2:
The patent adds a dimensional aspect by introducing class labels as an additional dimension in the feature representation. By incorporating class-specific representative vectors, the system creates a multi-dimensional feature space that preserves both robustness and class separability.
Data Source
AI summary
Calculate a plurality of feature vectors representing features of an input sample from the input sample which is multidimensional data by using a plurality of feature calculation models. Calculate similarity between an average value of the plurality of feature vectors and a representative vector corresponding to a class to which the input sample belongs among a plurality of representative vectors corresponding to a plurality of classes respectively, the representative vector having same dimensionality as each of the plurality of feature vectors. Learn parameters of the plurality of feature calculation models based on an evaluation function in which a value is larger as the similarity between the average value of the plurality of feature vectors and the representative vector corresponding to the class to which the input sample belongs is smaller.