Federated Learning Local Ground Truth Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing federated learning techniques inadequately create or identify ground truth data for updating machine learning models, especially in scenarios where such data is not readily available, leading to inefficiencies and inaccuracies in computer vision and other use cases.
Innovation Solution
The implementation of a method that utilizes locally-obtained ground truth data on user devices, leveraging a second prediction technique that differs from the first prediction technique being trained, to determine updates for the model, where the second prediction is considered more accurate and used as ground truth to compute gradients for updating the neural network, while ensuring user privacy through anonymous updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If user-provided ground truth data is used to assess predictions, then the model can be updated, but user privacy is compromised and ground truth data is not readily available in many use cases
Solution Approach 1:
The patent introduces an intermediary second prediction technique that acts as a mediator between the first prediction technique and the ground truth data. This intermediary generates synthetic ground truth data locally on user devices without requiring actual user-provided ground truth, thus enabling model updates while preserving user privacy and eliminating the need for sensitive data sharing.
Solution Approach 2:
The patent creates a copy of the ground truth data generation process by using a second prediction technique that replicates the functionality of the first prediction technique but with higher accuracy. This copy operates locally on user devices to generate synthetic ground truth labels without accessing actual user data, thereby enabling reliable model training while maintaining privacy.
2Measurement precision
If a second prediction technique is used to generate ground truth data, then ground truth accuracy is improved, but computational resources and processing time are increased
Solution Approach 1:
The patent applies partial action by using the second prediction technique selectively rather than continuously. The high-accuracy second technique is invoked only when ground truth data generation is needed for model updates, while the faster first prediction technique handles regular inference tasks. This partial application of computational resources ensures high ground truth accuracy when required without continuously consuming excessive computational energy.
3Ease of manufacture
If locally-obtained ground truth data is used, then model updates can be performed without sharing data, but creating accurate ground truth data becomes more complex
Solution Approach 1:
The patent makes the second prediction technique universal by designing it to serve multiple functions: it acts as both a prediction model and a ground truth data generator. This multi-functionality simplifies the overall system by eliminating the need for separate ground truth generation mechanisms, reducing device complexity while enabling private model updates through local ground truth creation.
Data Source
AI summary
Various implementations disclosed herein include devices, systems, and methods that involve federated learning techniques that utilize locally-determined ground truth data that may be used in addition to, or in the alternative to, user-provided ground truth data. Some implementations provide an improved federated learning technique that creates ground truth data on the user device using a second prediction technique that differs from a first prediction technique/model that is being trained. The second prediction technique may be better but may be less suited for real time, general use than the first prediction technique.