Knowledge aggregation networks for class incremental learning

Fu, Zhiling; Wang, Zhe; Xu, Xinlei; Li, Dongdong; Yang, Hai

doi:10.1016/j.patcog.2023.109310

Cited by 12 publications

(2 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to achieve a balance between model stability and plasticity during the training of the dual-model, we have introduced a loss function that considers three aspects: classification loss to address class imbalance, distillation loss to preserve old knowledge, and SupCon loss to learn new knowledge [75].…”

Section: ) Designing the Loss Functionmentioning

confidence: 99%

See 1 more Smart Citation

Privacy-Preserving Federated Class-Incremental Learning

Xiao,

Tang,

2024

Trans. Mach. Learn. Comm. Netw.

View full text Add to dashboard Cite

Federated Learning (FL) offers a collaborative training framework, aggregating model parameters from decentralized clients. Many existing models, however, assume static, predetermined data classes within FLa frequently unrealistic assumption. Real-time data additions from clients can degrade global model recognition of established classes due to catastrophic forgetting. This is exacerbated when new clients, unfamiliar to previous participants, join sporadically. Additionally, there's an imperative for client data privacy. Addressing these, we propose the Privacy-Preserving Federated Class-Incremental Learning (PP-FCIL) approach. This methodology ensures content-level privacy and significantly alleviates the risk of catastrophic forgetting in FCIL. To our knowledge, this is the first research seeks to embed differential privacy into the FCIL settings. Specifically, we introduce a dual-model structure that uses adaptive fusion of new and old knowledge to obtain a new global model. We also propose a multi-factor dynamic weighted aggregation strategy that considers several factors, such as data imbalance timeliness of the model, to speed up global model aggregation and accuracy. For privacy protection, we use Bayesian differential privacy to provide more balanced privacy protection for different datasets. Finally, we conducted experiments on CIFAR-100 and ImageNet to compare our method with other methods and verify its superiority.

show abstract

Section: ) Designing the Loss Functionmentioning

confidence: 99%

“…Similarly, F old demonstrates enhanced classification capabilities for images of old classes, given that Ω old is inherently designed to classify the old classes. To optimally harness the information from both F old and F new , we introduce the following dual-model adaptive feature fusion module [75]:…”

Section: ) Dual-model Adaptive Feature Fusionmentioning

confidence: 99%