ADCA: Attention-Driven Multi-Party Collusion Attack in Federated Self-Supervised Learning

Name: ADCA: Attention-Driven Multi-Party Collusion Attack in Federated Self-Supervised Learning
Uploaded: 2026-10-16T22:00:33.775630+00:00
Duration: 552 s
Description: The Attention-Driven multi-party Collusion Attack (ADCA) is a novel distributed backdoor attack for Federated Self-Supervised Learning (FSSL). It works by having malicious clients decompose a global trigger into local patterns and then collude. They use an attention mechanism to dynamically aggregate model updates, which helps hide the backdoor signal from benign client updates during training.

The Attention-Driven multi-party Collusion Attack (ADCA) is a novel distributed backdoor attack for Federated Self-Supervised Learning (FSSL). It works by having malicious clients decompose a global trigger into local patterns and then collude. They use an attention mechanism to dynamically aggregate model updates, which helps hide the backdoor signal from benign client updates during training.

Open the full episode