Publish time: 2024-09-09 00:00:00
32A-1130319
Patent pending
This research constructs innovative federated learning algorithms, delineating subspace through client model weight, thereby narrowing the optimal parameter search range. The accomplishments are listed as follows:
1. Extensively applicable to various neural network models and other machine learning algorithms.
2. Grounded in optimization theory, the complete algorithm is constructed through rigorous mathematical derivation.
3. Strengthening participant privacy and security protection by minimizing model parameter communication.
4. Optimization of the FedAvg algorithm, along with reducing system computational costs.
5. Targeting the FedAvg algorithm for enhancements, accelerating system convergence speed.
6. Adaptable to all scenarios of federated learning algorithms, minimizing system communication
costs.
7. Encouraging and evaluating participant contributions, proposing innovative incentive mechanisms and a fair business model framework and analysis.
- An innovative federated learning algorithm, where client model weights define a subspace, limiting optimal parameter search. Reduces communications frequency and volume, protecting individual privacy.
- Data contributors with high quality data are encouraged to participate through incentive mechanisms. The algorithm offers an optimal solution approximating FedAvg , reducing server and client computational resources.
Commercial sectors can benefit from this generalized federated learning framework, including finance, healthcare, biotechnology, telecommunications, education, and transportation. With intelligent recommendations, personalized advertising, and intricate customer analysis, it can enhance the user experience. In addition, federated learning is a critical component of smart cities and the Internet of Things.
Chiu Han Hsiao, Frank Yeong Sung Lin
If you are interested in this technology, please fill in the information on the webpage under the manufacturer option (Request for Technology Licensing, Material Transfer, or Sponsored Research) on the website of this office

Optimization based Eff icient and Effective Generic Federated Learning Algorithms and Fair Business Models