Xiang Song

Title: Xiang Song: Inventor of Language-Model Systems for Entity Representation

Introduction

Xiang Song is an inventor based in Los Altos, California, whose patent work explores how language models can combine different kinds of data to represent entities. His listed patent portfolio includes 1 patent, and he is associated with Amazon Technologies, Inc.

Latest Patents

Song’s patent, “Systems for Determining Representations of Entities Based on Graph Data Using a Language Model,” describes using a large language model (LLM) to generate an overall representation of an entity. The system combines representations derived from graph data, text, and images associated with that entity.

Graph data captures an entity’s characteristics and relationships with other entities, while text and image data provide additional information. These separate representations are supplied to an LLM trained to produce a unified entity representation. Other models can then use that representation to identify entities with similar or differing characteristics.

Career Highlights

The patent reflects work at the intersection of artificial intelligence, multimodal data processing, and entity representation. By bringing graph, text, and image information together, the described approach aims to provide a richer basis for downstream model tasks.

Collaborations

Song’s listed coworkers include Houyu Zhang and Seongjun Yun. The patent information provided describes a collaborative technical area but does not specify their individual contributions.

Conclusion

Xiang Song’s patent presents a method for combining graph, text, and image data through a language model to create useful entity representations. It highlights an approach to integrating diverse information for machine-learning applications.