Personal Machine Learning Model Segmentation for Edge Mobility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing personal machine learning models rely on pre-trained shared models, limiting mobility and privacy as they require large-scale data for training and cannot be easily transferred between devices, and personalization techniques do not allow for efficient incremental learning or deployment on constrained-resource devices.
Innovation Solution
A system comprising a task-independent shared ML model, a task-independent personal ML model, and a task-specific personal ML model, where personal models are trained using shared features and can be incrementally updated with user data, enabling deployment on edge devices and providing improved privacy and mobility by disentangling from pre-trained shared models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If personal machine learning models are trained using pre-trained shared models, then model accuracy is improved, but device mobility and privacy are worsened due to large-scale data requirements and inability to transfer between devices
Solution Approach 1:
The system segments the machine learning model into two distinct components: a shared model component trained on large-scale data for general accuracy, and a personal model component that can be independently trained and transferred between devices. This segmentation allows the personal model to be mobile while the shared model remains stationary on servers.
Solution Approach 2:
The patent extracts the personalization capability from the large pre-trained shared model, creating a separate personal model that contains only the essential personalized parameters. This extracted personal model can be transferred between devices without requiring the entire large-scale shared model, thus improving mobility while maintaining personalization benefits.
2Measurement precision
If personal machine learning models are trained using pre-trained shared models, then model accuracy is improved, but computational resource requirements are worsened making deployment on constrained devices difficult
Solution Approach 1:
The patent extracts only the necessary personal model parameters from the large shared model, creating a compact personal model that can run on constrained devices. This extracted personal model retains the ability to provide personalized accuracy while requiring minimal computational resources compared to deploying the full shared model on edge devices.
Solution Approach 2:
The system changes the parameters of the personal model by training it to learn user-specific patterns and preferences. This parameter optimization allows the small personal model to achieve high accuracy on constrained devices by focusing computational resources on learning only the essential personalized characteristics rather than general knowledge.
3Ease of operation
If personalization techniques are applied to pre-trained models, then user experience is improved, but incremental learning efficiency is worsened
Solution Approach 1:
The system segments the learning process into shared knowledge acquisition (performed once on servers) and personalization learning (performed incrementally on user devices). This segmentation allows the personal model to focus exclusively on incremental learning of user-specific patterns, improving learning efficiency while maintaining personalized user experience.
Solution Approach 2:
The shared model performs preliminary action by pre-training on large-scale data to establish general knowledge and features. This preliminary training removes the burden from the personal model, allowing it to focus only on incremental personalization learning, thereby improving incremental learning efficiency while still delivering personalized user experience.
Data Source
AI summary
Broadly speaking, embodiments of the present techniques provide a method and system for providing personal machine learning, ML, models. In particular, the present application provides a system for developing a training personal and personalised models to improve user experience.