Pre-computed memory or on-the-fly encoding? A hybrid approach to retrieval augmentation makes the most of your compute.
Michiel de JongYury ZemlyanskiyNicholas FitzGeraldJoshua AinslieSumit SanghaiFei ShaWilliam W. CohenPublished in: ICML (2023)
Keyphrases
- pre computed
- database
- information retrieval
- data sets
- document retrieval
- triangle inequality
- relevance feedback
- information retrieval systems
- retrieval accuracy
- associative memory
- retrieval model
- disk space
- index tree
- efficient retrieval
- image database
- main memory
- text retrieval
- data cube
- memory requirements
- retrieval systems
- content based retrieval
- bag of words
- indexing structure
- test collection
- memory space
- encoding scheme
- management system
- storage cost
- object recognition
- neural network