Glossary term
Glossary term
Architecture
A component of a deep neural network that is itself a deep neural network. In some cases, each tower reads from an independent data source, and those towers stay independent until their output is combined in a final layer. In other cases, (for example, in the encoder and decoder tower of many Transformers), towers have cross-connections to each other.
Created for this library
A retail recommender team uses a two-tower architecture with user and product towers to enable scalable embedding-based retrieval.
A search team uses a two-tower architecture so query and document embeddings are independent and can be indexed efficiently.
A research team uses a two-tower model with a shared embedding space for users and items in its candidate generator.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License