DCN-V2

What is DCN-V2? DCN-V2 is a type of architecture that is used in learning-to-rank. It is an improvement over the original DCN model. The main idea behind DCN-V2 is to learn explicit feature interactions through cross layers and combine them with a deep network to learn other implicit interactions. This architecture is capable of learning bounded-degree cross features. How Does DCN-V2 Work? The architecture of DCN-V2 involves two important components: explicit and implicit feature interaction