Representative Vector Generation for Memory-Constrained Identity Determination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep network methods for identity determination using representative vectors require significant memory usage due to the need for multiple representative vectors corresponding to each class, leading to increased memory requirements.
Innovation Solution
The proposed solution involves generating representative vectors by holding bases, including base vectors and weights for linear combination, which reduces the total memory size needed for learning, allowing for efficient generation and optimization of representative vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If representative vectors are stored directly for each class in deep network processing, then identity determination accuracy is maintained, but memory usage increases significantly
Solution Approach 1:
The representative vector is segmented into two components: a base vector (common to all classes) and a class-specific parameter. Instead of storing complete representative vectors for each class, the system stores the base vector once and individual parameters for each class, then reconstructs the representative vector during processing. This segmentation reduces memory usage while preserving the ability to perform accurate identity determination.
Solution Approach 2:
The base vector serves as a template that is copied and modified for each class by adding class-specific parameters. Rather than storing entirely separate representative vectors for each class, the system creates variations of the base vector through parameter addition, significantly reducing the total memory required while maintaining the functional equivalence of the original approach.
2Adaptability or versatility
If multiple representative vectors corresponding to each class are stored, then person class determination can be performed, but the total memory size increases
Solution Approach 1:
The base vector serves as a universal component that is shared across all classes. Instead of having separate, independent representative vectors for each class, the system uses a single base vector that can be combined with different class-specific parameters to generate representative vectors for any class. This multi-functional approach maintains class determination capability while reducing memory requirements.
Solution Approach 2:
The system merges the common features (base vector) and class-specific features (parameters) into a unified representation framework. By combining these two elements, the system achieves the functionality of storing multiple representative vectors while actually storing far less data, as the base vector is shared across all classes and only the smaller parameters need to be stored individually.
Data Source
AI summary
An information processing apparatus includes one or more memories, and one or more processors that, when executing instructions stored in the one or more memories, function as the following units: an acquisition unit configured to acquire learning data including data and a label indicating a category of the data, a base holding unit configured to hold a base for generating a representative vector in the category, a learning unit configured to learn a parameter related to generation of the representative vector based on the acquired learning data, and a first generation unit configured to generate the representative vector based on the parameter and the base.