Kexin Zhao

Title: Innovations by Kexin Zhao

Introduction

Kexin Zhao is an accomplished inventor based in Santa Clara, CA. He holds a total of 3 patents that focus on advancements in neural networks and audio processing technologies. His work has significantly contributed to the fields of machine learning and audio synthesis.

Latest Patents

Kexin Zhao's latest patents include innovative systems and methods for trace norm regularization and faster inference for embedded models. These patents describe techniques for compressing and accelerating dense matrix multiplications, particularly in neural networks used for large vocabulary continuous speech recognition (LVCSR). The introduction of trace norm regularization techniques allows for better accuracy in low-rank training, enhancing the training speed of large models. Additionally, his work on small-footprint flow-based models for raw audio, known as WaveFlow, showcases a generative flow that efficiently synthesizes high-fidelity speech. WaveFlow utilizes a dilated two-dimensional convolutional architecture to manage long-range waveform structures while maintaining a compact model size of 5.91M parameters, making it significantly smaller than existing models.

Career Highlights

Kexin Zhao is currently employed at Baidu USA LLC, where he continues to push the boundaries of technology in audio and machine learning. His innovative approaches have garnered attention in the tech community, particularly for their practical applications in real-time audio generation.

Collaborations

Kexin has collaborated with notable colleagues, including Kainan Peng and Wei Ping, who contribute to the dynamic research environment at Baidu USA LLC.

Conclusion

Kexin Zhao's contributions to the fields of neural networks and audio processing exemplify the spirit of innovation. His patents not only advance technology but also pave the way for future developments in these critical areas.