Junfeng Wen

Title: Junfeng Wen and Collaborative Private-Data Model Training.

Introduction

Junfeng Wen is an inventor based in Waterloo, Canada. His patent work includes a method for collaborative model training that allows systems to learn from private data sets without revealing the data themselves.

Latest Patents

Wen is associated with 1 patent: “Distributed Model Training with Collaboration Weights for Private Data Sets.” The invention enables model-training systems to collaborate while keeping their respective data sets private. Each data set learns client weights for a collection of computer models, and inference for that data set combines model parameters according to those weights.

During training, the client weights are updated based on how well sampled models represent each private data set. The system also calculates gradients for sampled models, which can be weighted by the corresponding client weights. This approach increases the influence of a data set on model parameters that are more relevant to it.

Career Highlights

Wen is associated with The Toronto-Dominion Bank. His patent reflects work on collaborative machine learning, with an emphasis on training across private data sets while limiting disclosure of sensitive information.

Collaborations

Wen’s listed coworkers include Jesse Cole Cresswell and Brendan Leigh Ross.

Conclusion

Junfeng Wen’s patent presents a collaborative approach to model training that uses client-specific weights to learn from private data sets without exposing the underlying data.