Venkata S Bhojanapalli

Title: Venkata S. Bhojanapalli and Efficient Transformer Attention

Introduction

Venkata S. Bhojanapalli is an inventor associated with Google Inc. in New York, NY (US). His patent work addresses ways to make attention-based machine-learning models more computationally efficient.

Latest Patents

Bhojanapalli is associated with 1 patent, titled “Leveraging Redundancy in Attention with Reuse Transformers.” The invention describes systems and methods for reusing attention scores across layers or heads in Transformers and other attention-based neural networks. By carrying forward scores computed in one layer to one or more subsequent layers, the proposed approach aims to reduce the computational cost of self-attention while maintaining comparable or potentially better results.

Career Highlights

The available information connects Bhojanapalli’s patent activity with Google Inc. and highlights his work on improving the efficiency of Transformer architectures. The patent focuses on exploiting redundancy in attention computations to reduce repeated processing.

Collaborations

The patent’s listed collaborators include Andreas Veit and Ayan Chakrabarti, reflecting a collaborative effort on attention reuse in machine-learning models.

Conclusion

Venkata S. Bhojanapalli’s patent presents an approach to reducing the computational demands of Transformer models by reusing attention scores, contributing to ongoing innovation in efficient machine learning.