Hee Woo Jun

Title: Hee Woo Jun: Innovator in Deep Learning and Audio Synthesis

Introduction

Hee Woo Jun is a prominent inventor based in Sunnyvale, CA, known for her contributions to deep learning and audio synthesis technologies. With a total of 4 patents, she has made significant strides in understanding the complexities of model accuracy and computational scaling in deep learning applications.

Latest Patents

One of her latest patents focuses on "Predicting Deep Learning Scaling." This work emphasizes the importance of understanding the relationships between training set size, computational scale, and model accuracy improvements. The patent presents a large-scale empirical study of error and model size growth as training sets increase. It introduces methodologies for measuring these relationships and predicting other metrics, highlighting that power-law can represent deep model relationships effectively. The findings have substantial implications for deep learning research, model debugging, and computing system design.

Another notable patent is "Spectrogram to Waveform Synthesis Using Convolutional Networks." This invention addresses the challenge of waveform synthesis from spectrograms by presenting an efficient neural network architecture based on transposed convolutions. The architecture achieves high compute intensity and fast inference, yielding high-quality audio synthesis that is over 500 times faster than real-time audio. The model also utilizes a GAN framework to enhance audio quality, demonstrating significant improvements over traditional iterative algorithms.

Career Highlights

Hee Woo Jun is currently employed at Baidu USA LLC, where she continues to innovate in the fields of deep learning and audio processing. Her work has garnered attention for its practical applications and contributions to advancing technology in these areas.

Collaborations

Hee Woo Jun has collaborated with notable colleagues, including Gregory Diamos and Joel Hestness, further enriching her research and development efforts.

Conclusion

Hee Woo Jun's innovative work in deep learning and audio synthesis showcases her expertise and commitment to advancing technology. Her patents reflect a deep understanding of complex relationships in computational models and their practical applications.