Xin Li

Title: Innovations of Xin Li in Few-Shot Class-Incremental Learning

Introduction

Xin Li is an innovative inventor based in Sunnyvale, CA. He has made significant contributions to the field of machine learning, particularly in few-shot class-incremental learning. His work focuses on enhancing the performance of image-text classification tasks through a minimalist approach.

Latest Patents

Xin Li holds a patent titled "Minimalist multi-modal approach to few-shot class-incremental learning." This patent describes methods and systems for Few-Shot Class-Incremental Learning (FSCIL) that utilize a combination of Session Specific Prompts (SSP) and hyperbolic distance metrics. These techniques enhance session-wise learning and representation of image-text pairings across differing classes. The methods include a base training session where both text and image features are projected into hyperbolic space for accurate class pairing using a cross-entropy loss function. Subsequent incremental sessions incorporate previously learned SSPs to retain and augment the separability of classes while minimizing the trainable parameters. This innovative approach achieves higher accuracy with fewer trainable parameters compared to traditional models.

Career Highlights

Xin Li is currently employed at Robert Bosch, where he continues to develop and refine his innovative ideas. His work has garnered attention for its potential to revolutionize the way machine learning models are trained and utilized.

Collaborations

Xin collaborates with talented individuals such as Thang Doan and Sima Behpour, who contribute to the advancement of their projects and research initiatives.

Conclusion

Xin Li's contributions to few-shot class-incremental learning demonstrate his commitment to innovation in machine learning. His patent reflects a significant advancement in the field, showcasing the potential for improved accuracy and efficiency in image-text classification tasks.