Peng Dai

Title: Peng Dai and Asynchronous Federated Learning

Introduction

Peng Dai is an inventor based in Mountain View, California. His patent work concerns federated learning, a method for training machine-learning models across multiple institutions while coordinating updates to a shared global model.

Latest Patents

Dai is associated with 1 patent: “System and Method for Asynchronous Multi-Aspect Weighted Federated Learning.” The invention describes a system for asynchronously training a federated learning model using a master and multiple institutional clients. The master runs an aggregation thread and separate client-management threads. Client threads initiate local training or continue training after receiving an aggregation event, then send training events to the aggregation thread. The aggregation thread uses those events to update the global model and sends aggregation events back to the relevant client threads.

Career Highlights

Dai’s listed patent work reflects an interest in coordinating distributed model training. The system’s asynchronous design allows client training and global-model aggregation to proceed through exchanges of training and aggregation events.

Collaborations

Dai is associated with Beijing Wodong Tianjun Information Technology Co., Ltd. His listed coworkers include Huasong Shan and Tao Zeng.

Conclusion

Peng Dai’s patent describes an asynchronous approach to federated learning that coordinates local training across institutional clients with updates to a shared global model.