On-Device AI Model Personalization Without Data Storage Growth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training artificial intelligence models require large storage capacity, increased computer resources, and suffer from catastrophic forgetting when updating models with new data, especially in personalization scenarios where user-specific data is involved.
Innovation Solution
Implementing personal and general training data generation models on-device to generate training data locally, allowing for incremental model updates without extensive storage or network connection, and using unsupervised learning to maintain model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is continuously accumulated to improve AI model performance, then model accuracy is improved, but storage capacity requirements increase
Solution Approach 1:
The patent creates synthetic training data by training a generative model (copier) to replicate the characteristics of accumulated training data. Instead of storing all original training data, the system uses the generative model to generate new training samples on-demand, reducing storage requirements while maintaining model accuracy through synthetic data replication
Solution Approach 2:
The patent extracts only the essential characteristics and patterns from the accumulated training data by training a compact generative model. This extracted knowledge is stored in a condensed form (model parameters) rather than retaining the full volume of original training data, achieving storage efficiency while preserving the core information needed for model training
2Measurement precision
If more training data is accumulated to improve personalization, then user-specific accuracy is improved, but computer resources for learning increase
Solution Approach 1:
The patent uses a generative model to generate only the necessary portion of training data needed for personalization rather than processing or storing all available training data. This partial action approach generates sufficient synthetic samples for effective model training while avoiding the excessive computational burden of handling complete training datasets
Solution Approach 2:
The system performs preliminary training of the generative model offline or in advance, extracting and storing the essential patterns of user-specific data beforehand. This preliminary action prepares the generative model to quickly generate personalized training data without requiring heavy computational resources during actual model training operations
3Measurement precision
If training data volume increases to improve model performance, then model accuracy is improved, but update time increases
Solution Approach 1:
The system creates a compressed representation (generative model) that copies the essential characteristics of large training datasets. This copied knowledge can be rapidly instantiated to generate training samples without the time-consuming process of loading and processing voluminous original data, significantly reducing model update time while maintaining accuracy
Solution Approach 2:
The patent transforms the training data from a large-volume state to a compact parameter representation through the generative model. By changing the state of data storage from raw samples to model parameters, the system enables faster data generation and model training operations, reducing update time while preserving the information needed for accurate training
4Measurement precision
If existing training data is replaced with new data to enable personalization, then user-specific performance is improved, but previously learned knowledge is forgotten
Solution Approach 1:
The patent merges general training data with user-specific training data by training the generative model on a combination of both data types. This merging approach allows the system to learn from diverse sources simultaneously, enabling personalization while preserving previously learned general knowledge through the integrated training process
Solution Approach 2:
The system dynamically adjusts the composition of training data by using the generative model to blend general and user-specific data in appropriate proportions. This dynamic approach allows the model to adaptively incorporate new user-specific information while maintaining previously learned general knowledge, preventing catastrophic forgetting through flexible data composition
Data Source
Figure 1~3
Figure 4~5
Figure 6~7
AI summary
Disclosed is an electronic apparatus. The electronic apparatus may include a memory configured to store one or more training data generation models and an artificial intelligence model, and a processor configured to generate personal training data that reflects a characteristic of a user using the one or more training data generation models, train the artificial intelligence model using the personal learning data as training data, and store the trained artificial intelligence model in the memory.