LITTA: Late-Interaction and Test-Time Alignment for Visually-Grounded Multimodal Retrieval

1. 1Generate complementary queries to expand the user's query
2. 2Align the query and document representations at test time
3. 3Use late interaction to improve the retrieval of relevant evidence pages
4. 4Evaluate the performance of LITTA on multimodal document retrieval tasks

Researchers and engineers working on multimodal retrieval and question-answering systems can benefit from LITTA, as it enhances the retrieval of relevant evidence from visually rich documents

💡 LITTA's query-expansion-centric approach and test-time alignment enable effective retrieval of relevant evidence pages from visually rich documents

📚 LITTA: a new framework for multimodal retrieval that improves document retrieval without retraining #multimodalretrieval #questionanswering