Wearable Inference Model Update via Gradient Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Wearable apparatuses face challenges in accurately classifying images while preserving user privacy, as existing systems often result in erroneous classifications and require sharing of personal data to correct these errors.
Innovation Solution
A server-based system determines a global update to the inference model by analyzing update information from multiple wearable devices, including directions of gradients derived from captured images, and distributes this update to preserve user privacy and improve image classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image data is shared to correct classification errors, then classification accuracy is improved, but user privacy is compromised
Solution Approach 1:
The patent extracts only the essential gradient information needed for model improvement while leaving out the actual image data. The gradient directions are derived from image data but contain insufficient information to reconstruct the original images, thus separating the useful signal from the privacy-sensitive content.
Solution Approach 2:
Instead of sharing image data to improve the model, the patent inverts the approach by sharing gradient information derived from the images. This reversal allows model improvement while inherently protecting privacy, as the gradients alone cannot reveal the original images.
2Reliability
If gradient information is transmitted for model updates, then classification errors are corrected, but data transmission requirements increase
Solution Approach 1:
The patent extracts only the necessary gradient direction information from complete image data, transmitting only what is needed for model updates. This extraction reduces the data volume significantly compared to transmitting full images while maintaining the ability to correct classification errors.
3Adaptability or versatility
If multiple devices contribute to model updates, then model robustness is improved, but system complexity increases
Solution Approach 1:
The patent implements a universal gradient aggregation mechanism that can handle contributions from multiple wearable devices with the same inference model. The server aggregates gradient information from various devices to improve the global model, making the system multi-functional in handling diverse device inputs through a unified process.
Data Source
AI summary
A system and a method determine a global update to an inference model and distribute the global update to a plurality of wearable apparatuses. In one implementation, the system includes a data interface and at least one processing device. The at least one processing device is programmed to receive, via the data interface and from a first one of the plurality of wearable apparatuses, first update information associated with the inference model. The at least one processing device is further programmed to receive, via the data interface and from a second one of the plurality of wearable apparatuses, second update information associated with the inference model. The at least one processing device is further programmed to analyze the first update information and the second update information to determine the global update to the inference model, and transmit, via the data interface, the global update to the plurality of wearable apparatuses.