On-Device Machine Learning Model Update Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large machine learning models, such as those used in speech recognition and natural language processing, face challenges in efficient updates on-device due to their size and bandwidth constraints, leading to potential reductions in accuracy during over-the-air (OTA) updates.
Innovation Solution
A system and method for updating on-device machine learning models using a trained model update object, which is smaller than the model itself, allowing for offline updates without interfering with user activities, and can be compressed to a higher degree without accuracy loss, using techniques like sparse matrices, hashed representations, or input-output pairs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the entire machine learning model is transmitted for OTA updates, then the model can be fully updated, but the data transfer size becomes excessively large
Solution Approach 1:
The patent extracts only the essential update information from the complete model, creating a condensed model update object that contains only the necessary parameters and configurations needed to transform the previous model into the updated model. This extraction process significantly reduces data transfer size while preserving the essential update functionality.
Solution Approach 2:
Instead of transmitting the entire updated model, the patent creates a simplified copy or representation (the model update object) that captures the essential transformation information. This update object serves as a compressed copy that can be transmitted efficiently and then applied to generate the full updated model on the target device.
2Quantity of substance
If the model is compressed to reduce data transfer size, then bandwidth requirements are reduced, but model accuracy may be compromised
Solution Approach 1:
The patent changes the parameters of the model representation by transforming the complete model into a different parameter space (the model update object parameters). This parameter transformation allows the model to be represented in a more compact form that can be efficiently transmitted while preserving the essential information needed to reconstruct the updated model with full accuracy.
3Productivity
If OTA updates are performed during user activities, then updates can be applied in real-time, but user experience is degraded due to interference
Solution Approach 1:
The patent performs the model update operation offline, before the user needs to use the updated functionality. The model update object is transmitted and stored in advance, and the actual model transformation is performed when the device is not actively being used, ensuring that updates are ready before users need them without interfering with user activities.
4Reliability
If the complete model is transmitted, then no accuracy loss occurs, but bandwidth consumption increases significantly
Solution Approach 1:
The patent extracts only the essential update information from the complete model, creating a condensed model update object that contains only the necessary parameters and configurations needed to transform the previous model into the updated model. This extraction process significantly reduces data transfer size while preserving the essential update functionality.
Data Source
AI summary
Some devices may perform processing using machine learning models trained at a centralized system and distributed to the device. The centralized system may update the machine learning model and distribute the update to the device (or devices). To reduce the size of an update, the centralized system may train a model update object, which may be smaller in size than the model itself and thus more suitable for sending to the device(s). A device may receive the model update object and use it to update the on-device machine learning model; for example, by changing some parameters of the model. Parameters left unchanged during the update may retain their previous value. Thus, using the model update object to update the on-device model may result in a more accurate updated model when compared to sending an updated model compressed to a size similar to that of the model update object.