Chenghui Wei

works on

backdoor defense 1 federated learning 1 gradient projection 1 language model fine-tuning 1 optimizer state entanglement 1 privacy attacks 1

From the 1 of 1 linked paper with an AI index.

collaborators

Showing cs.LGShow all

1 paper · 1 filter

cs.LG2026

TriShield: Zero-Utility-Loss Defense Against Privacy Backdoors in Federated Language Model Fine-Tuning via Orthogonal Gradient Projection and Optimizer State Entanglement

Cheng Wei

The paper introduces TriShield, a three‑component defense that eliminates privacy backdoor reconstruction attacks in federated fine‑tuning of large language models without any loss…