Researchers have developed CACO, a new self-supervised post-training framework designed to enhance continuous generative retrieval systems. This framework optimizes query coordination to improve item coverage and recall without significantly increasing inference latency. CACO achieves this by coordinating queries from existing flow-matching and diffusion models, using empirical retrieval frequencies as a proxy for relevance to create a label-free coverage objective. Experiments show CACO can improve recall by up to 23.9% and reduce inter-query overlap by over 50% across various benchmarks. AI
IMPACT Enhances retrieval systems by improving recall and reducing redundancy with minimal latency overhead.
RANK_REASON The cluster contains a research paper detailing a new method for generative retrieval. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
- alphaXiv
- artificial neural network
- arXiv
- CACO
- CatalyzeX
- DagsHub
- diffusion
- Flow Matching for Generative Modeling
- Gotit.pub
- Hugging Face
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →