On-Device Machine Learning Platform for Secure Local Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning approaches require transmitting data over networks, leading to security risks, increased network traffic, reduced communication speeds, and latency, while on-device implementations are resource-intensive and complicate application development.
Innovation Solution
An on-device machine learning platform that performs prediction, training, and example collection locally, using a centralized service to manage machine-learned models and context features, enabling secure and efficient machine learning operations without direct network transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are implemented on user devices, then data security is improved and network traffic is reduced, but device resources are consumed and application complexity increases
Solution Approach 1:
The patent introduces an on-device machine learning platform as an intermediary layer between applications and machine learning models. This platform provides prediction, training, and example collection services to multiple applications, thereby reducing individual application complexity while maintaining data security on-device. The platform acts as a mediator that handles the intricacies of machine learning operations centrally.
Solution Approach 2:
The on-device machine learning platform is designed to serve multiple applications with diverse machine learning needs through a unified architecture. It provides universal services including prediction generation, model training, and example collection that can be accessed by any application on the device, thereby reducing overall system complexity while maintaining security.
2Speed
If machine learning models are stored and implemented within applications, then inference speed is improved, but application size and memory footprint increase
Solution Approach 1:
The patent merges the machine learning capabilities of multiple applications into a single on-device platform. Instead of each application storing its own models separately, the platform consolidates models and serves them to multiple applications, reducing redundant storage while maintaining fast local inference performance.
Solution Approach 2:
The patent segments the machine learning functionality into a separate on-device platform that is independent of individual applications. This segmentation allows applications to access machine learning services without embedding the actual models, thereby reducing application size while maintaining inference speed through local processing.
3Measurement precision
If training examples are collected and transmitted to centralized servers, then model accuracy is improved, but network traffic and latency increase
Solution Approach 1:
The patent implements preliminary action by performing model training on-device using collected examples before deployment. This allows the model to be trained locally with high accuracy while minimizing the need for frequent data transmission to centralized servers, thereby reducing training latency and network dependency.
Solution Approach 2:
The on-device machine learning platform enables self-service by autonomously collecting training examples, training models, and updating predictions locally without requiring constant communication with centralized servers. This self-sufficient approach maintains model accuracy while eliminating network transmission delays.
4Adaptability or versatility
If applications include their own machine learning services, then service functionality is improved, but development complexity and maintenance burden increase
Solution Approach 1:
The patent introduces an on-device machine learning platform as an intermediary service that applications can access through standardized interfaces. This mediator provides comprehensive machine learning functionality including prediction, training, and example collection, thereby maintaining service versatility while simplifying application development by abstracting away the complexity of implementing these services directly in applications.
Data Source
AI summary
The present disclosure provides systems and methods for on-device machine learning. In particular, the present disclosure is directed to an on-device machine learning platform and associated techniques that enable on-device prediction, training, example collection, and/or other machine learning tasks or functionality. The on-device machine learning platform can include a context provider that securely injects context features into collected training examples and/or client-provided input data used to generate predictions/inferences. Thus, the on-device machine learning platform can enable centralized training example collection, model training, and usage of machine-learned models as a service to applications or other clients.