FedHAN: A Cache-Based Semi-Asynchronous Federated Learning Framework Defending Against Poisoning Attacks in Heterogeneous Clients

Xiaoding Wang, Li Xu, Bin Ye, Lizhao Wu, Sun-Yuan Hsieh, Jie Wu, Limei Lin

Abstract:

Federated learning is vulnerable to model poison
ing attacks in which malicious participants com
promise the global model by altering the model up
dates. Current defense strategies are divided into
three types: aggregation-based methods, validation
dataset-based methods, and update distance-based
methods. However, these techniques often neglect
the challenges posed by device heterogeneity and
asynchronous communication. Even upon identi
fying malicious clients, the global model may al
ready be significantly damaged, requiring effec
tive recovery strategies to reduce the attacker’s im
pact. Current recovery methods, which are based
on historical update records, are limited in en
vironments with device heterogeneity and asyn
chronous communication. To address these prob
lems, we introduce FedHAN, a reliable federated
learning algorithm designed for asynchronous com
munication and device heterogeneity. FedHAN
customizes sparse models, uses historical client up
dates to impute missing parameters in sparse up
dates, dynamically assigns adaptive weights, and
combines update deviation detection with update
prediction-based model recovery. Theoretical anal
ysis indicates that FedHAN achieves favorable con
vergence despite unbounded staleness and effec
tively discriminates between benign and malicious
clients. Experiments reveal that FedHAN, com
pared to leading methods, increases the accuracy
of the model by 7.86%, improves the detection ac
curacy of poisoning attacks by 12%, and enhances
the recovery accuracy by 7.26%. As evidenced by
these results, FedHAN exhibits enhanced reliability
and robustness in intricate and dynamic federated
learning scenarios.