Back to search

University of Washington

Adversarial Trained Deep Learning Poisoning Defense: SpaceTime

Abstract

dc:description.abstract

Poisoning defenses for federated learning are in- creasingly needed to harden security solutions for the Internet of Things (IoT). This study aimed to examine defensive mech- anisms to protect these federated learning models from the growing vulnerability presented by poisoning attacks. Poison- ing attacks may use Byzantine or Sybil methodologies in a centralized or distributed deployment. The attacks may use label-flipping or backdoor vectors to affect the global training model. In physics, spacetime is a mathematical model that combines the three dimensions of space and the one dimension of time into a four dimensional manifold. Poisoning attacks have complex spatial and time relationships that present identi- fiable patterns in that manifold. This state-of-the-art poisoning defense was built upon a time series regression many-to- one architecture using spacetime relationships to provide an adversarial trained deep learning poisoning defense. Earlier algorithms have utilized different spatial similarity metrics from Euclidean Distance (ED), cosine similarity (CS), and other pairwise measurements to determine poisoning attacks. Different algorithms have provided better defenses against specific attacks while also exhibiting unique vulnerabilities. We suggest SpaceTime – Deep Similarity Defense, a deep learning recurrent neural network that includes space and time perceptions to provide a defense against poisoning attacks for federated learning models. Our evaluation shows that SpaceTime exceeds the current research models for Byzan- tine and Sybil attacks using label flipping, backdoor, and distributed backdoor methodologies. Our results include non- independent and identical participant data and various attack methodologies.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Dunham, Christian
Advisor dc:contributor.advisor
  • Thamilarasu, Geetha

Subjects

dc:subject × 8

Rights

dc:rights
Statement dc:rights
  • CC BY-NC-SA
Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/1773/49650

Chain of custody

source
Harvested from
University of Washington
Base URL
digital.lib.washington.edu/server/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Dunham, Christian. Adversarial Trained Deep Learning Poisoning Defense: SpaceTime. 2023. http://hdl.handle.net/1773/49650