Michal Lukasik

Title: Michal Lukasik: Inventor Advancing Transformer Efficiency.

Introduction

Michal Lukasik is an inventor associated with New York, NY (US), whose patent work addresses computational efficiency in attention-based machine-learning models. His disclosed invention focuses on reusing attention scores across Transformer layers and heads.

Latest Patents

Lukasik is associated with 1 patent: “Leveraging Redundancy in Attention with Reuse Transformers.” The patent describes systems and methods that reuse attention scores computed in one layer in one or more subsequent layers. This approach aims to reduce the computational cost of self-attention models while maintaining comparable or potentially superior results.

Career Highlights

Lukasik works at Google Inc. His patent reflects work on improving the efficiency of Transformer architectures by taking advantage of redundancy in attention computations.

Collaborations

The patent lists Venkata S. Bhojanapalli and Andreas Veit among Lukasik’s coworkers, indicating collaboration with researchers working on machine-learning systems and architectures.

Conclusion

Michal Lukasik’s patent contribution centers on reusing attention scores to make Transformer models more computationally efficient.