Research talk: Is phrase retrieval all we need?

Speaker: Danqi Chen, Assistant Professor, Princeton University DensePhrases is an extractive phrase-search tool based on natural language input that achieves dense retrieval of billion-scale phrases with extreme runtime efficiency. In this talk, Assistant Professor Danqi Chen of Princeton University will highlight some of the technical challenges that she and her research team encountered and the solutions of learning dense representations of phrases at scale. She’ll demonstrate the strong performance on open-domain QA and slot-filling tasks, and she’ll show how phrase retrieval, the most fine-grained retrieval unit, can also be used for passage or document retrieval tasks. Finally, she’ll cover how phrase filtering and vector quantization can make the phrase index much smaller, making dense phrase retrieval a practical and versatile solution in multi-granularity retrieval. Learn more about the 2021 Microsoft Research Summit: