Collaborative ML Classification via Device Selection Server
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Machine Learning (ML) technologies on mobile devices face limitations in processing power, data storage, and battery life, making them inefficient for standalone ML tasks, prompting the need for collaborative classification methods among multiple devices within a network.
Innovation Solution
A communications device acquires a feature vector representing an instance, classifies it using a local ML model, and if the confidence level is below a threshold, it requests assistance from other devices based on user identity, contact lists, data type, origin, location, and related instances, to enhance classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If Machine Learning is performed on mobile devices, then classification capability is provided, but processing power requirements exceed device limitations
Solution Approach 1:
The system segments the classification task by first performing a preliminary classification on the mobile device using a lightweight local ML model. If the confidence level is insufficient, the system then segments the remaining complex classification task and offloads it to a server, allowing the mobile device to handle simple tasks independently while relying on server power for complex tasks.
Solution Approach 2:
The patent introduces a server as an intermediary component that receives classification requests from mobile devices, performs complex ML classification using its powerful processing resources, and returns results to the mobile devices. This intermediary enables mobile devices to access advanced classification capabilities without requiring them to have the necessary processing power.
2Measurement precision
If collaborative classification is implemented, then classification accuracy is improved, but device complexity increases
Solution Approach 1:
The mobile device autonomously performs the following steps without requiring complex coordination: it locally classifies the instance using its ML model, evaluates the confidence level, and only if insufficient, requests assistance from the server. This self-service approach simplifies the device's role while improving accuracy through collaborative processing.
Solution Approach 2:
The system uses feedback mechanisms where the mobile device receives the classification result and confidence level from the server, and this feedback information is used to update the local ML model. This feedback loop enables continuous improvement of classification accuracy while keeping the device logic relatively simple.
3Speed
If local ML models are used, then processing speed is improved, but model accuracy is limited by device capabilities
Solution Approach 1:
The mobile device performs partial classification action by using a simplified local ML model that provides quick results. When the confidence level indicates insufficient accuracy, the system supplements this partial action by requesting more accurate classification from the server, thus achieving both speed and accuracy through a combination of local and remote processing.
Data Source
AI summary
A selection server for selecting one or more other communications devices for classifying an instance using Machine Learning, ML, is provided. The selection server is operative to receive, from a communications device for classifying an instance using ML, a selection request message for selecting one or more other communications devices for classifying an instance using ML, the selection request message comprising information pertaining to at least one of: an identity of a user of the communications device, a contact list of the user, a type of data comprised in a feature vector representing the instance, an origin of the feature vector, a classification of the instance using a local first ML model of the communications device, a location of the communications device, a location associated with the instance, and one or more classified instances which are related to the instance represented by the feature vector.