Long Thanh Le

Title: Long Thanh Le and Machine-Learned Model Alignment

Introduction

Long Thanh Le is an inventor based in Sunnyvale, California, and works at Google Inc. His patent work concerns adapting machine-learned models to improve their ability to follow instructions.

Latest Patents

Le is associated with 1 patent, titled “Machine-Learned Model Alignment with Synthetic Data.” The invention describes computer-implemented systems and methods for adapting machine-learned models with high-quality synthetic data tailored to particular instruction distributions and target models.

The disclosed system uses metadata about a task’s use case and required skills to guide a generative model in creating a synthetic instruction and a corresponding response. A target sequence-processing model can then be modified using that instruction-response pair, with the aim of improving its instruction-following abilities.

Career Highlights

Le’s listed patent highlights work on synthetic-data-driven model adaptation. The patent describes a process for conditioning generated training examples on task metadata, connecting specific use cases and skills with model training.

Collaborations

Le’s listed coworkers include Zifeng Wang and Vincent Perot. The patent information provided describes the technology but does not specify their roles in its development.

Conclusion

Long Thanh Le’s patent work at Google Inc. focuses on using metadata-conditioned synthetic data to adapt machine-learned models and strengthen their instruction-following performance.