Soyoung Peraud

Title: Soyoung Peraud: Innovator in Classifier Assistance Technology

Introduction

Soyoung Peraud is a notable inventor based in Redmond, WA, who has made significant contributions to the field of machine learning and data classification. With a focus on enhancing classifier performance through innovative techniques, she has developed a unique approach that leverages domain-trained embeddings.

Latest Patents

Soyoung holds a patent titled "Classifier assistance using domain-trained embedding." This invention allows classifiers to be trained with a limited number of manually annotated datasets. By utilizing a small subset of verbatims labeled with topic labels as seeds, her method employs data augmentations to create seed verbatim sets for known topics. The process involves assigning temporary pseudo labels to the remaining verbatims based on their proximity in vector space to the labeled seeds. The training consists of classification epochs that update embeddings, assuming pseudo labels are ground-truth labels, and labeling epochs that refine the vectors corresponding to the verbatims. As the training progresses, the embeddings converge, leading to improved classification accuracy.

Career Highlights

Soyoung is currently associated with Microsoft Technology Licensing, LLC, where she applies her expertise in machine learning and data analysis. Her work focuses on developing innovative solutions that enhance the efficiency and effectiveness of classification systems.

Collaborations

Throughout her career, Soyoung has collaborated with talented individuals such as Alexandre Rochette and Gabriel Arien Desgarennes. These partnerships have contributed to the advancement of her research and the successful implementation of her inventions.

Conclusion

Soyoung Peraud is a pioneering inventor whose work in classifier assistance technology is shaping the future of machine learning. Her innovative approaches and collaborations continue to drive advancements in the field, making her a significant figure in the realm of data classification.