Decentralized Model Parameter Extraction for Privacy and Bandwidth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for updating learned models in data analytics, particularly those using sensed data from portable devices, face challenges in maintaining user privacy and efficiency, as they often require sending raw data to external servers, which can be insecure and bandwidth-intensive, and may not provide high-quality analytics due to limited processing power on user devices.
Innovation Solution
A decentralized system where local learned models on user devices generate local parameters from raw data, combine them to create global parameters, and send only these parameters to a remote server for updating, allowing for secure, bandwidth-efficient updates of models without exposing raw user data, while also enabling personalized and aggregated analytics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw data is sent to external servers for model training, then model training quality is improved, but user data security and bandwidth efficiency deteriorate
Solution Approach 1:
The patent extracts only the essential model parameters from raw data for transmission to external servers, leaving the sensitive raw data local to users. This extraction approach enables model training without compromising user data security, as only anonymized parameter updates are transmitted rather than complete raw datasets.
Solution Approach 2:
The patent introduces local learned models as intermediaries that process raw data locally and generate parameter updates. These intermediate parameter representations serve as mediators between raw user data and external training servers, enabling quality model training while maintaining data security through the intermediary transformation layer.
2Measurement precision
If raw data is sent to external servers for model training, then model training quality is improved, but bandwidth consumption increases
Solution Approach 1:
The patent extracts only the essential model parameters from raw data for transmission to external servers, leaving the sensitive raw data local to users. This extraction approach enables model training without compromising user data security, as only anonymized parameter updates are transmitted rather than complete raw datasets.
Solution Approach 2:
The patent transforms raw data into parameter representations through local learned models before transmission. This parameter transformation significantly reduces data volume and bandwidth consumption while preserving the essential information needed for model training, as parameters are compact mathematical representations compared to raw sensed data.
3Reliability
If local learned models are updated using only local parameters, then user privacy is improved, but analytics quality deteriorates
Solution Approach 1:
The patent merges local parameters with updated global parameters received from external servers to update local learned models. This combination enables the model to benefit from both local user data patterns (maintaining privacy) and global knowledge from aggregated data (improving analytics quality), creating a synergistic effect that resolves the contradiction between privacy and quality.
Solution Approach 2:
The patent implements a feedback mechanism where updated global parameters are transmitted back to local devices and integrated with local parameters. This feedback loop allows local models to continuously improve their analytics quality by incorporating global insights while maintaining the privacy benefits of local processing, creating a continuous improvement cycle.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods and systems are disclosed for updating learned models. An embodiment comprises receiving a plurality of data sets representing sensed data from one or more devices and determining, using one or more local learned models, local parameters based on the received data sets. Another operation may comprise generating a combined data set by combining the plurality of data sets and, determining, using one or more local learned models, global parameters based on the combined data set. Another operation may comprise transmitting, to a remote system, the global parameters for determining updated global parameters using one or more global learned models based at least partially on the global parameters, and receiving, from the remote system, the updated global parameters. Another operation may comprise updating the one or more local learned models using both the local parameters and updated global parameters.