Lijin Yang

Title: Lijin Yang and Video Extraction Model Generation

Introduction

Lijin Yang is an inventor based in Koto-ku, Japan. His listed patent work concerns machine learning methods for generating a model that identifies video intervals matching the content of an input sentence.

Latest Patents

Yang is associated with one patent, 1 patent, titled “Model Generation Method and Model Generation System.” The invention describes training a video extraction model to find intervals in a video that correspond to an input sentence.

During training, the model extracts a base matching interval and a sub matching interval from a training video using a base sentence and a sub sentence. It then calculates losses for the ground-truth interval and the two extracted intervals by attempting to reconstruct the base sentence from video features associated with each interval. The model is trained so that the loss for the ground-truth interval is lower than the loss for the base matching interval, which is in turn lower than the loss for the sub matching interval.

Career Highlights

Yang is associated with Toyota Jidosha Kabushiki Kaisha, commonly known as Toyota Motor Corporation. His listed patent illustrates work at the intersection of video understanding and machine learning.

Collaborations

Hsuan-Kung Yang and Norimasa Kobori are identified as Yang’s coworkers. The available information does not specify their roles in the patent or the extent of their collaboration.

Conclusion

Lijin Yang’s listed invention focuses on training a model to locate video segments that match sentence content, reflecting an application of machine learning to video retrieval.