Hye Kang Park

Title: Hye Kang Park and Confidence Calibration Innovation.

Introduction

Hye Kang Park is an inventor based in Seoul, South Korea, associated with Yonsei University. Her patent work addresses confidence calibration in image classification, with a focus on adapting label smoothing to the difficulty of classification tasks.

Latest Patents

Park is associated with 1 patent, titled “Apparatus and Method for Confidence Calibration Based on Adaptive Label Smoothing.” The invention describes a computing-device method that inputs image data into an image classification model and obtains the model’s logits and prediction probabilities. It then produces a label-smoothing loss function using learnable margin parameters, label-smoothing weights based on image-classification difficulty, or both.

Career Highlights

Park’s patent highlights an approach to calibrating model confidence by adapting label smoothing to the classification context. The disclosed method connects image difficulty with the loss function used during model training.

Collaborations

The patent information names Bum Sub Ham and Jong Youn Noh as Park’s coworkers. Her listed institutional affiliation is Yonsei University.

Conclusion

Hye Kang Park’s patent presents an adaptive label-smoothing method for confidence calibration in image classification, reflecting her work associated with Yonsei University.