Enrique Alfonseca

Title: Enrique Alfonseca: Inventor of an Efficient Dual-Encoder Retrieval Architecture.

Introduction

Enrique Alfonseca is an inventor based in Mountain View, California, and works at Google Inc. His patent work addresses the efficiency of retrieval systems that use dual-encoder architectures.

Latest Patents

Alfonseca is associated with 1 patent: “Dual Encoder Retrieval Efficiency with Parameter Sharing in Projection Layer.” The invention describes an asymmetric dual-encoder architecture with separate token-embedding and encoder sections for two inputs. A shared projection layer processes encodings from both sections to generate projections for an embedding space. That space produces question and answer embeddings, which can be used to identify candidate answers to an input query.

Career Highlights

The patent reflects work on improving retrieval efficiency through parameter sharing in the projection layer. It combines distinct processing paths for the two inputs with a shared component for generating their projections.

Collaborations

The patent lists Zhe Dong and Jianmo Ni among Alfonseca’s coworkers.

Conclusion

Enrique Alfonseca’s patent describes a dual-encoder retrieval approach that uses a shared projection layer to support the generation of question and answer embeddings.