Autoregressive Retriever (ARR)
What it is
ARR is a multimodal retrieval model that iteratively refines query embeddings using retrieved items as feedback. It combines supervised fine-tuning with reinforcement learning to optimize the selection of informative items for better ranking.
The problem it solves
Standard retrieval methods encode queries once and fail to update the query representation even when retrieved items could help clarify the user's intent.
What you could use it for
- Universal multimodal retrieval systems
- Improving zero-shot retrieval performance
- Query understanding in search engines
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.