Machine Learning Model Reconstruction via Erasure Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models trained and stored on user devices are susceptible to accidental deletion or corruption, leading to the need for time-consuming retraining, which can disrupt device functionality and expose sensitive user data.
Innovation Solution
The model parameters are divided into smaller sets and encoded using erasure coding, allowing reconstruction with only a portion of the sets, distributing these across secondary devices for enhanced security and rapid recovery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are trained and stored on user devices for security reasons, then data privacy is improved, but the models become highly susceptible to accidental deletion or corruption
Solution Approach 1:
The model parameters are divided into multiple portions and distributed across different devices. Each device stores only a fragment of the complete model, so accidental deletion at any single device does not result in complete data loss. The model can be reconstructed by gathering sufficient portions from multiple devices.
Solution Approach 2:
Erasure coding is applied in advance to encode the model parameters into multiple redundant portions before distribution. This preliminary encoding ensures that even if some portions are lost or corrupted, the original model can be recovered as long as a sufficient number of portions are available.
2Reliability
If complete model parameter sets are distributed across multiple devices, then security is improved, but storage requirements increase
Solution Approach 1:
The system discards the need to store complete model copies at each device by using erasure coding. Instead, each device stores only essential portions, and the complete model can be recovered by combining sufficient portions from multiple devices, thereby reducing redundant storage.
3Object-affected harmful factors
If model parameters are divided and distributed across secondary devices, then security against deletion is improved, but model reconstruction complexity increases
Solution Approach 1:
The patent replaces complex manual model reconstruction processes with automated erasure decoding algorithms. The system automatically retrieves model portions from distributed devices, decodes them using erasure coding algorithms, and reconstructs the complete model without requiring manual intervention or complex coordination.
Data Source
AI summary
A method performed by one or more processors that preserves a machine learning model comprises accessing model parameters associated with a machine learning model. The model parameters are determined responsive to training the machine learning model. The method comprises generating a plurality of model parameter sets, where each of the plurality of model parameter sets comprises a separate portion of the set of model parameters. The method comprises determining one or more parity sets comprising values calculated from the plurality of model parameter sets. The method comprises distributing the plurality of model parameter sets and the one or more parity sets among a plurality of computing devices, where each of the plurality of computing devices stores a model parameter set of the plurality of model parameter sets or a parity set of the one or more parity sets. The method comprises accessing, from the plurality of computing devices, a number of sets comprising model parameter sets and at least one parity set. The method comprises reconstructing the machine learning model from the number of sets accessed from the plurality of computing devices.