Xue Lin

Title: Innovations of Xue Lin in Deep Neural Network Optimization

Introduction

Xue Lin is an accomplished inventor based in Newton, MA (US). He has made significant contributions to the field of deep learning through his innovative patent. His work focuses on optimizing deep neural network models, which is crucial for enhancing computational efficiency and performance.

Latest Patents

Xue Lin holds a patent for "Computer-implemented methods and systems for compressing deep neural network models using alternating direction method of multipliers (ADMM)." This patent introduces ADMM-NN, an algorithm-hardware co-optimization framework for deep neural networks (DNNs) utilizing the Alternating Direction Method of Multipliers (ADMM). The framework consists of two main components: a systematic joint framework for DNN weight pruning and quantization, and a hardware-aware optimization to facilitate hardware-level implementations. The ADMM-based weight pruning and quantization lead to significant computation reduction and energy efficiency improvements. Experimental results indicate that by combining weight pruning and quantization, the proposed framework achieves a remarkable 1,910× reduction in overall model size for the LeNet-5 model and 231× for the AlexNet model. Additionally, a 3.6× reduction in computation can be achieved without any loss in accuracy, surpassing previous methodologies.

Career Highlights

Xue Lin is affiliated with Northeastern University, where he continues to advance research in deep learning and neural network optimization. His innovative approaches have garnered attention in the academic and technological communities, contributing to the evolution of machine learning applications.

Collaborations

Xue Lin collaborates with Yanzhi Wang, a fellow researcher who brings valuable insights and expertise to their joint projects. Their partnership enhances the research output and fosters innovation in the field of deep learning.

Conclusion

Xue Lin's contributions to deep neural network optimization through his patented innovations demonstrate his commitment to advancing technology. His work not only improves computational efficiency but also sets a new standard for future research in the field.