Homomorphic Encryption for Federated Learning Model Confidentiality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated learning of artificial neural networks faces challenges in maintaining confidentiality of model parameters, as the aggregation platform can access and potentially reverse-engineer the training data from the updated models, which is a security concern for data suppliers.
Innovation Solution
Implementing homomorphic encryption to encrypt model parameters during transmission and processing, allowing data suppliers to update models locally and combine them securely without sharing intermediate or final models with the aggregation platform, ensuring only encrypted parameters are exchanged, and using a shared private key for decryption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If federated learning is implemented without encryption, then collaborative learning efficiency is improved, but model parameter confidentiality deteriorates
Solution Approach 1:
Homomorphic encryption acts as an intermediary mechanism that allows the aggregation platform to process model parameters without being able to read their actual values. The encryption scheme enables mathematical operations on encrypted data, serving as a mediator between the need for centralized aggregation and the requirement for parameter confidentiality.
Solution Approach 2:
The patent transforms model parameters from their original plaintext form to encrypted form before transmission and processing. This parameter transformation maintains the functional capability of the learning system while fundamentally changing the security properties of the data being exchanged.
2Speed
If model parameters are transmitted in plaintext for aggregation, then processing speed is improved, but data security deteriorates
Solution Approach 1:
Homomorphic encryption serves as an intermediary that enables fast aggregation processing while maintaining security. The encryption scheme is designed to allow efficient mathematical operations on encrypted values, preventing a security-speed tradeoff.
3Manufacturing precision
If data suppliers share intermediate models with the platform, then model convergence is improved, but information leakage risk increases
Solution Approach 1:
The patent applies encryption transformation to model parameters at all stages of the federated learning process, including intermediate models. This ensures that even when models are repeatedly exchanged between suppliers and the platform for convergence, the confidentiality property is maintained throughout the entire process.
Data Source
AI summary
A method for federative learning of an artificial neural network model on a plurality of sets of training data. The learning method used a plurality of data suppliers each having a distinct set of training data and a symmetric key, and an aggregation platform aggregating partial models trained on a sub-plurality of these sets, in each iteration. In each iteration, the platform selects a sub-plurality of data suppliers and supplies the parameters of the model to them, in homomorphically encrypted form. Each training data supplier decrypts these parameters, trains the model on his own data set and returns the parameters of the partial model thus obtained, encrypted by his symmetric key, to the platform. The aggregation platform then transcrypts these partial models in the homomorphic domain, and combines them in the homomorphic domain to obtain a global model.