App to App Retrieval
App-to-app retrieval focuses on efficiently and accurately retrieving relevant information from large datasets, particularly within the context of augmented generation models. Current research emphasizes improving retrieval methods through techniques like contrastive learning, recursive abstractive processing, and the incorporation of derived features or semantic similarity searches, often within hierarchical memory structures. These advancements are crucial for enhancing various applications, including fact-checking, question answering, recommendation systems, and robotics, by enabling more effective use of large language models and multimodal data. The ultimate goal is to create more robust and efficient systems capable of handling increasingly complex information retrieval tasks.
Papers
Remember, Retrieve and Generate: Understanding Infinite Visual Concepts as Your Personalized Assistant
Haoran Hao, Jiaming Han, Changsheng Li, Yu-Feng Li, Xiangyu Yue
Probing-RAG: Self-Probing to Guide Language Models in Selective Document Retrieval
Ingeol Baek, Hwan Chang, Byeongjeong Kim, Jimin Lee, Hwanhee Lee
CrediRAG: Network-Augmented Credibility-Based Retrieval for Misinformation Detection in Reddit
Ashwin Ram, Yigit Ege Bayiz, Arash Amini, Mustafa Munir, Radu Marculescu
MultiVENT 2.0: A Massive Multilingual Benchmark for Event-Centric Video Retrieval
Reno Kriz, Kate Sanders, David Etter, Kenton Murray, Cameron Carpenter, Kelly Van Ochten, Hannah Recknor, Jimena Guallar-Blasco, Alexander Martin, Ronald Colaianni, Nolan King, Eugene Yang, Benjamin Van Durme