Collaborative Machine Learning for IoT Label Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Individual devices with low processing performance face limitations in building trained models that can cover diverse input data, as they are unable to process large datasets and have a limited label coverage area, making it difficult to apply machine learning effectively in IoT environments.
Innovation Solution
A machine learning system and method that enables cooperation between multiple devices to extend the coverage area of a trained model, where a first learning device determines the cluster for input data using a first artificial neural network and transmits sample features to a second learning device with more processing power, which determines the label using a second artificial neural network, allowing for improved AI performance without exposing personal information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If an individual device with low processing performance performs machine learning on its own, then the device can operate independently, but the label coverage area is greatly limited and cannot cover diverse input data
Solution Approach 1:
The patent combines multiple learning devices to form a collaborative machine learning system. Devices with different processing capabilities work together, where each device contributes its strengths. The low-power device performs initial processing while the high-power device handles complex analysis, achieving extended label coverage without requiring every device to have high processing power independently.
Solution Approach 2:
The patent introduces a hierarchical dimension to machine learning operations. Instead of a single processing level, it creates multiple processing tiers: local processing on individual devices and centralized processing on high-power devices. This dimensional change allows the system to overcome the processing power limitations of individual low-power devices while maintaining their operational independence.
2Productivity
If a trained model is built on an individual device, then the device can make local decisions, but it is unable to process large datasets and the model coverage is limited
Solution Approach 1:
The patent segments the machine learning workload into different components distributed across multiple devices. Low-power devices handle data collection, preliminary processing, and simple inference tasks. High-power devices handle large-scale data processing, complex model training, and generate comprehensive trained models that are then distributed back to all devices. This segmentation allows each device to operate within its capability limits while collectively achieving high productivity.
3Adaptability or versatility
If multiple devices cooperate to extend model coverage, then the label coverage area increases, but communication overhead and information exposure risks increase
Solution Approach 1:
The patent extracts and transmits only essential features and aggregated data between devices, rather than sharing complete raw datasets. Each device extracts relevant feature representations from its local data and shares only these processed features with other devices. This extraction approach enables collaborative model training and coverage extension while minimizing the exposure of personal and sensitive information contained in the original data.
Data Source
AI summary
Disclosed is an artificial intelligence or machine learning algorithm that may be applied to a plurality of machine learning devices in a 5G environment connected to perform the Internet of things. A machine learning method by a first learning machine according to one embodiment of the present disclosure may include obtaining input data; determining, from among a plurality of clusters, a cluster to which the input data belongs, by using a first artificial neural network; transmitting a plurality of sample features associated with the determined cluster to a second learning device using a second artificial neural network; receiving a label for the plurality of sample features from the second learning device, in response to the transmission; and associating the received label with the determined cluster.