Boosting Protocols on Encrypted Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face inefficiencies in applying homomorphic encryption to complex problems, such as training deep neural networks, due to computational constraints and the need to protect private data and intellectual property.
Innovation Solution
The method involves training machine-learning models using simpler models like logistic regression and boosting protocols on homomorphically encrypted data, ensuring that client data remains encrypted and cloud platform models are not revealed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If homomorphic encryption is applied to complex machine-learning problems, then data privacy is protected, but computational efficiency deteriorates
Solution Approach 1:
The patent segments the machine-learning training process into multiple iterative boosting rounds, where each round processes encrypted data independently. This segmentation allows the system to maintain strong privacy protection through homomorphic encryption while improving computational efficiency by breaking down the complex training task into manageable iterations with intermediate results that can be processed incrementally
Solution Approach 2:
The patent applies preliminary action by pre-processing the encrypted data and transforming it into a format suitable for boosting algorithms before the main training process. This includes preparing encrypted feature representations and initial model states in advance, which reduces the computational burden during the actual training iterations and improves overall efficiency without compromising privacy
2Productivity
If simpler models like logistic regression are used with boosting protocols, then training efficiency improves, but model complexity and capability are limited
Solution Approach 1:
The patent merges multiple simple logistic regression models through the boosting protocol to create a powerful ensemble model. Each individual model remains computationally simple and efficient to train, but their combination through iterative boosting produces a complex, high-capability model that can solve sophisticated machine-learning problems while maintaining training efficiency at the component level
Solution Approach 2:
The patent transitions from a single complex model in one dimension to multiple simple models stacked in the iteration dimension. By adding the time/iteration dimension to the training process, the system achieves the capability of complex models while maintaining the simplicity and efficiency of basic logistic regression units, effectively trading spatial complexity for temporal progression
Data Source
AI summary
In an approach for training machine-learning models using encrypted data, a processor receives a set of encrypted data from a client computing device. A processor trains a machine-learning model using a boosting algorithm. A processor performs a first classification on the set of encrypted data using the machine-learning model. A processor sends a first set of encrypted results of the first classification to the client computing device. A processor receives a first set of boosting updates from the client computing device. A processor applies the first set of boosting updates to the machine-learning model.