betaFeedback

WorksComputer Science

Stealing Exact Training Data from Other Clients during Federated Learning

Abstract

Malicious adversaries can employ gradient inversion attacks to reverse-engineer private data using the model updates of a round in Federated Learning. While there have been developments in gradient inversion for text, exact text recovery by a malicious client has so far remained infeasible, as FedAvg entangles the gradient contributions of many clients and obscures the signal of any individual sentence. We introduce a novel attack pipeline that unwinds this entanglement progressively through an iterative loop, in which the gradient contributions of sentences recovered exactly in one round are subtracted from the aggregate, exposing the remaining sentences for recovery in the next iteration. We show, for the first time, exact sentence recovery, with 100% precision, and on aggregates containing up to 4×4\times more tokens than the model's hidden dimension, we recover up to 80.2% of client sentences exactly, a regime where existing attacks recover none exactly.

Paper details

vlnc:W_yje8ex25
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
  • Computer Science

LLMs, AI-based coding agents

View statement
Looking for related papers…