Siddharth Goyal

Title: Siddharth Goyal: Innovator in Neural Network Optimization

Introduction

Siddharth Goyal is a prominent inventor based in San Jose, California. He has made significant contributions to the field of neural networks, particularly in optimizing matrix multiplications for embedded models. His work is crucial for enhancing the performance of large vocabulary continuous speech recognition systems.

Latest Patents

Siddharth holds a patent titled "Systems and methods for trace norm regularization and faster inference for embedded models." This patent describes innovative systems and methods for compressing and accelerating dense matrix multiplications, which are essential in fully connected and recurrent layers of neural networks. The techniques introduced in this patent utilize trace norm regularization for training low rank factored versions of matrix multiplications. Compared to standard low rank training, these methods consistently achieve better accuracy while balancing the number of parameters. Additionally, the patent discusses how faster inference can be achieved on ARM processors through optimized kernels for small batch sizes, resulting in significant speed improvements over existing libraries. These techniques are not only applicable to large vocabulary continuous speech recognition but also to embedded neural networks with substantial fully connected or recurrent layers.

Career Highlights

Siddharth Goyal is currently employed at Baidu USA LLC, where he continues to push the boundaries of innovation in neural network technology. His work has been instrumental in advancing the capabilities of embedded systems, making them more efficient and effective.

Collaborations

Siddharth collaborates with talented individuals such as Kexin Zhao and Mohammad Shoeybi. Their combined expertise contributes to the development of cutting-edge technologies in the field of artificial intelligence and machine learning.

Conclusion

Siddharth Goyal is a key figure in the innovation of neural network optimization, with a focus on enhancing the efficiency of embedded models. His contributions through patents and collaborations are paving the way for advancements in artificial intelligence applications.