Jian Meng

Title: Jian Meng and Temperature-Resilient Neural Network Training

Introduction

Jian Meng is an inventor associated with Arizona State University in Tempe, Arizona. His patent work addresses a challenge in neural-network training: maintaining model performance across changes in operating temperature.

Latest Patents

Meng is named on 1 patent, titled “Method and System for a Temperature-Resilient Neural Network Training Model.” The invention describes loading a neural-network model onto a resistive nonvolatile in-memory-computing chip and training it with progressive knowledge distillation. The process first uses a clean teacher model to introduce low-temperature noise into a student model, then uses the student as the teacher while increasing the noise to represent higher temperatures. The method also adapts batch-normalization parameters to account for multiple thermal variations.

Career Highlights

The patent reflects work at the intersection of neural-network training, in-memory computing, and temperature resilience. Its approach combines staged noise injection with batch-normalization adaptation to help a model accommodate thermal changes.

Collaborations

The patent’s listed collaborators include Jae-sun Seo and Li Yang. Meng is associated with Arizona State University.

Conclusion

Jian Meng’s patent presents a training approach designed to improve neural-network resilience to temperature variation in resistive in-memory-computing systems.