Qijun Luo

Title: Qijun Luo and Deep Convolutional Neural Network Normalization

Introduction

Qijun Luo is an inventor based in Shenzhen, China, whose patent work addresses training methods for deep convolutional neural networks. His listed patent concerns normalization techniques intended to support stable performance across different training batch sizes and vision tasks.

Latest Patents

Luo is associated with 1 patent: “Normalization in Deep Convolutional Neural Networks.” The invention describes a machine-learning device with a first neural network layer, a second neural network layer, and a normalization layer between them. During training, the normalization layer receives outputs for multiple samples in a batch, groups data values according to indices across two dimensions, and produces normalized outputs for the second layer. The approach aims to improve training performance and generalizability across vision tasks while maintaining stability at different batch sizes.

Career Highlights

Luo is associated with Huawei Technologies Co., Ltd. His patent reflects work on neural network training and normalization, with a focus on improving the reliability and performance of deep learning systems.

Collaborations

The provided information lists Xiaoyun Zhou and Jiacheng Sun as Luo’s coworkers. Their specific roles in the patent or related work are not specified.

Conclusion

Qijun Luo’s patent focuses on a normalization method designed to make deep convolutional neural network training more stable and adaptable across batch sizes and vision tasks.