Ka Ho Yenson Lau

Title: Ka Ho Yenson Lau: Innovator in Distributed Model Training

Introduction

Ka Ho Yenson Lau is a notable inventor based in Toronto, Canada. He has made significant contributions to the field of model training systems, particularly in the area of private data sets. His innovative approach allows for collaboration in model training without compromising the privacy of individual data sets.

Latest Patents

Ka Ho Yenson Lau holds a patent for "Distributed model training with collaboration weights for private data sets." This patent describes a system where model training occurs collaboratively while keeping respective private data sets confidential. Each private data set learns a set of client weights for a series of computer models that are also developed during the training process. The inference for a specific private data set is determined as a mixture of the computer model parameters based on the client weights. During each training iteration, the client weights are updated based on how well the sampled models represent the private data set. Additionally, gradients are determined for each sampled model and may be adjusted according to the client weight for that model, thereby enhancing the gradient contribution of a private data set for model parameters that are more closely aligned with it.

Career Highlights

Ka Ho Yenson Lau is currently employed at the Toronto-Dominion Bank, where he applies his expertise in model training systems. His work focuses on enhancing the efficiency and privacy of data handling in financial applications.

Collaborations

He collaborates with talented individuals such as Junfeng Wen and Jesse Cole Cresswell, contributing to innovative projects that push the boundaries of technology in data privacy and model training.

Conclusion

Ka Ho Yenson Lau is a pioneering inventor whose work in distributed model training is shaping the future of data privacy in collaborative systems. His contributions are vital in ensuring that private data sets can be utilized effectively without compromising confidentiality.