Mohsen Gholami

Title: Mohsen Gholami and Knowledge Transfer Between Language Models.

Introduction

Mohsen Gholami is an inventor associated with Huawei Cloud Computing Technologies in Vancouver, Canada. His patent work addresses a challenge in artificial intelligence: transferring knowledge from large language models to smaller models.

Latest Patents

Gholami is credited with 1 patent: “Methods and Systems for Transferring Knowledge from Large Language Models to Small Language Models Using Out-of-Distribution Feedbacks.” The invention describes an iterative process in which a large language model generates training samples for a small language model, which is then trained to perform a specified task. The large model also generates validation samples, and selected validation data is added to the training set for a subsequent iteration. This approach is designed to use feedback beyond the small model’s initial data samples to guide its training.

Career Highlights

Gholami’s listed patent reflects work on model training and knowledge transfer. The disclosed method combines generated training examples, validation, and iterative refinement to support the development of small language models.

Collaborations

The provided information lists Mohammad Akbari and Vaden Wybert Masrani as Gholami’s coworkers. The patent description also identifies collaboration between large and small language models as central to the invention’s approach.

Conclusion

Mohsen Gholami’s patent focuses on an iterative method for transferring knowledge from large language models to small language models, with validation feedback helping shape later training rounds.