Storage Architecture for Compressed Deep Learning Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning models face issues such as insufficient learning data, imbalance of data classes, and overfitting, which are not adequately addressed by existing data augmentation techniques.
Innovation Solution
A storage system and method that compresses augmented data units generated from original data units, storing them with alteration information units, and decompresses them during learning operations to restore the augmented data units for use in host systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is performed to generate more learning data, then the quantity of learning data is increased, but the storage space required increases proportionally
Solution Approach 1:
The patent extracts and stores only the transformation parameters (alteration information) that define how augmented data is generated from original data, rather than storing the complete augmented data units. This separation allows the system to retain the ability to regenerate augmented data on-demand while minimizing storage requirements.
Solution Approach 2:
The system pre-calculates and stores the alteration information (transformation parameters) during data preparation, enabling rapid regeneration of augmented data during learning operations without performing computationally intensive augmentation transformations in real-time.
2Productivity
If all augmented data units are stored in advance, then the learning operation speed is improved, but the storage space requirement increases
Solution Approach 1:
Instead of storing multiple copies of augmented data, the system stores a single copy of the original data unit along with transformation parameters that enable virtual copying of augmented versions when needed, significantly reducing storage requirements while maintaining access efficiency.
Solution Approach 2:
The alteration information acts as an intermediary between the original data and the augmented data, enabling the system to generate augmented data on-demand without storing all possible augmented versions, thus balancing storage efficiency with access speed.
3Volume of stationary object
If compression is applied to reduce storage space, then the storage efficiency is improved, but the decompression time increases
Solution Approach 1:
The system transforms augmented data into a compressed parameter representation (alteration information) that encodes the transformation operations needed to reconstruct the augmented data. This parameter-based compression achieves high compression ratios while enabling rapid regeneration through simple parameter application rather than lengthy decompression processes.
Data Source
AI summary
A controller included in a storage system which performs based on an original data unit, a compression operation on each of augmented data units to generate one or more alteration information units respectively corresponding to the augmented data units; and which stores original data unit and the alteration information units in the storage device.