Jie Meng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jie Meng has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- Zhuang Yang
Soochow University
2 shared publications - Qiong Wu
Department of Psychology, Suzhou University of Science and Technology, Suzhou, China
1 shared publication - Wencong Jia
Soochow University
1 shared publication - Xi Chen
School of Software, Soochow University, Suzhou, Jiangsu, China
1 shared publication - Xizhao Luo
School of Computer Science and Technology, Soochow University, Suzhou, Jiangsu, China
1 shared publication
Latest Publications
- 18 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-11
Fast Federated Learning Frameworks, Driven by Powerball Function and Hypergradient Descent
Federated learning (FL), a paradigm of distributed machine learning, aims to utilize user data for model training while safeguarding data privacy. However, numerous challenges arise in practical applications. Among these, accelerating the training rate and improving the generalization performance of FL models are particularly noteworthy challenges. Recent studies are particularly about understanding the role of stochastic optimization in federated optimization, leading to various fast federated optimization algorithms. As a special case, some studies have explored treating the sum of local differences as pseudo-gradients and also introduced the global learning rate. To deal with the challenge, this work develops fast federated optimization algorithms from the perspective of the Powerball function and the hypergradient descent, coined FedPBHD, extending the family of federated optimization algorithms. Different from existing FL algorithms, FedPBHD renews the update direction by using the discretization’s of ordinary differential equations (ODEs). Besides, FedPBHD utilizes an adaptive update rule of the learning rate (an extremely crucial hyperparameter for FL algorithms). The convergence performance of FedPBHD has been established through a rigorous theoretical analysis. Finally, we compare our FedPBHD algorithms against numerous benchmark optimizers for different practical applications of computer vision, natural language processing under both independent identically distribution (IID) and non-IID conditions. Experimental results on different problems have shown encouraging performance of the novel Federated optimization algorithms.
- 18 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-11
Fast Federated Learning Frameworks, Driven by Powerball Function and Hypergradient Descent
Federated learning (FL), a paradigm of distributed machine learning, aims to utilize user data for model training while safeguarding data privacy. However, numerous challenges arise in practical applications. Among these, accelerating the training rate and improving the generalization performance of FL models are particularly noteworthy challenges. Recent studies are particularly about understanding the role of stochastic optimization in federated optimization, leading to various fast federated optimization algorithms. As a special case, some studies have explored treating the sum of local differences as pseudo-gradients and also introduced the global learning rate. To deal with the challenge, this work develops fast federated optimization algorithms from the perspective of the Powerball function and the hypergradient descent, coined FedPBHD, extending the family of federated optimization algorithms. Different from existing FL algorithms, FedPBHD renews the update direction by using the discretization’s of ordinary differential equations (ODEs). Besides, FedPBHD utilizes an adaptive update rule of the learning rate (an extremely crucial hyperparameter for FL algorithms). The convergence performance of FedPBHD has been established through a rigorous theoretical analysis. Finally, we compare our FedPBHD algorithms against numerous benchmark optimizers for different practical applications of computer vision, natural language processing under both independent identically distribution (IID) and non-IID conditions. Experimental results on different problems have shown encouraging performance of the novel Federated optimization algorithms.
- 9 January 2026
- Published by Springer Nature
- p. 212-225
Lightweight Diffusion Models with Federated Learning: Attention No Necessary for U-Net
- 9 January 2026
- Published by Springer Nature
- p. 212-225
Lightweight Diffusion Models with Federated Learning: Attention No Necessary for U-Net