Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Robustness federated learning algorithm based on partial parameter aggregation

A technology of parameter aggregation and learning algorithm, which is applied in computing, computer parts, digital data protection, etc., can solve the problem that the server is difficult to verify the correctness of users, and achieve the effects of weakening attack capabilities, improving robustness, and ensuring data privacy

Active Publication Date: 2021-08-06
NANKAI UNIV
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, due to the application of secure aggregation algorithms in federated learning, it is difficult for the server to verify the correctness of the results uploaded by users

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Robustness federated learning algorithm based on partial parameter aggregation
  • Robustness federated learning algorithm based on partial parameter aggregation
  • Robustness federated learning algorithm based on partial parameter aggregation

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0026] In order to explain in detail the technical content, structural features, achieved goals and effects of the technical solution, the following will be described in detail in conjunction with specific embodiments and accompanying drawings.

[0027] The present invention proposes a robust federated learning algorithm based on partial parameter aggregation. First, the server defines a unified upload ratio for each client, and distributes it to the client together with the global model. After calculating the update of the local model, the client selects parameters in the model that meet the upload ratio, which effectively reduces the model information uploaded by malicious clients, but still ensures the correct convergence of the global model. Then, based on homomorphic encryption, the present invention designs encrypted calculations for some of the models uploaded by the client, so that the server can still only obtain the aggregated results of the model parameters, but cann...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention belongs to the technical field of federated learning robustness, and particularly relates to a robustness federated learning algorithm based on partial parameter aggregation, which comprises a basic partial aggregation protocol and a security partial aggregation algorithm based on homomorphic encryption. Aiming at the problem that a server or a third-party mechanism is difficult to detect malicious users and is difficult to resist backdoor attacks from a client in a federated learning training scene, a partial aggregation protocol is designed, the capability of attacking the users by malicious backdoors is limited while stable convergence of a model is ensured, the robustness of a federated learning system is remarkably enhanced, and the invention is especially suitable for large-scale user joint training scenes. Meanwhile, in order to ensure privacy of data and models participating in training of clients, a security aggregation algorithm based on homomorphic encryption is designed for the aggregation algorithm of the part, and it is ensured that data uploaded by a user is invisible to a server. Therefore, the security of federal learning on the client side and the server side is ensured.

Description

technical field [0001] The present invention belongs to the research in the field of federated learning robustness, and specifically relates to a robust federated learning algorithm based on partial parameter aggregation, aiming at federated learning including a partial federated learning aggregation algorithm (PartialFedAvgalgorithm) and a secure aggregation encryption protocol based on partial aggregation (Partial Secure Aggregation Protocol). Background technique [0002] Federated Learning technology provides a security solution for massive end-user cooperative training models. Federated learning technology allows users to upload model parameters instead of directly uploading private data. At the same time, it is guaranteed that any uploaded data of the user is under the encryption protection of the security aggregation algorithm, which further protects the privacy of the user's data. In the federated learning process, the server first initializes a global model and di...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06F21/55G06F21/60G06K9/62
CPCG06F21/602G06F21/55G06F18/214
Inventor 刘哲理侯博禹高继强郭晓杰张宝磊
Owner NANKAI UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products