Guolong Su

Title: Guolong Su: Inventor in Sequence Modeling

Introduction

Guolong Su is an inventor based in State College, Pennsylvania. His work includes a patent on structure-aware sequence modeling, a field concerned with how machine-learning systems interpret and process text.

Latest Patents

Su is associated with 1 patent, titled “Structural Encoding and Attention Paradigms for Sequence Modeling.” The patent describes systems and methods for interpreting a document’s text without first determining its correct reading order. In some examples, a graph convolutional network generates contextualized “supertoken” embeddings, which are then processed by a transformer using sparse attention. The attention weights for some supertokens are adjusted according to differences between predicted and actual order and distance values.

Career Highlights

Su works at Google Inc. His patent reflects an approach to sequence modeling designed to handle documents whose reading order may not be straightforward.

Collaborations

Su’s coworker Chun-Liang Li is also associated with the work.

Conclusion

Guolong Su’s patent presents a structure-aware approach to document interpretation that combines graph-based contextual embeddings with transformer attention.