Research talk: Learning and pretraining strategies for dense retrieval in search and beyond

Subscribers:
351,000
Published on ● Video Link: https://www.youtube.com/watch?v=Ewyre3RQhSo



Duration: 10:28
1,698 views
0


Speaker: Chenyan Xiong, Senior Researcher, Microsoft Research Redmond

In this talk, we’ll quickly go through our recent observations and findings with dense retrieval. First, we’ll recap the standard setup and training strategies for dense retrieval models in search. We’ll then share our recent progress on generalizing dense retrieval to recommendation, complex question answering, and latent reasoning. Finally, we’ll discuss the challenges with mismatches between pretrained models and dense retrieval requirements and our early attempts in dense retrieval dedicated pretraining techniques.

Learn more about the 2021 Microsoft Research Summit: https://Aka.ms/researchsummit




Other Videos By Microsoft Research


2022-02-08Research talk: Getting to net zero safely
2022-02-08Keynote: Building a net-zero future together
2022-02-08Opening Remarks: Research for Carbon Negative
2022-02-08Research talk: Computationally efficient large-scale AI
2022-02-08Research talk: Towards data-efficient machine learning with meta-learning
2022-02-08Research talk: Resource-efficient learning for large pretrained models
2022-02-08Research talk: Transformer efficiency: From model compression to training acceleration
2022-02-08Plenary: Industrial Research in the 21st Century
2022-02-02Dead-end Discovery: How offline reinforcement learning could assist healthcare decision-makers
2022-01-27Microsoft Soundscape - overview of Routes feature
2022-01-24Research talk: Learning and pretraining strategies for dense retrieval in search and beyond
2022-01-24Keynote: Accelerating the data and AI transformation of industry
2022-01-24Research talk: Revisiting data center management: Moving towards self-managed data center networks
2022-01-24Fireside chat: Network verification: Cloud first, what's next?
2022-01-24Research talk: Successor feature sets: Generalizing successor representations across policies
2022-01-24Plenary: The Future of Research
2022-01-24Keynote: Universal search and recommendation
2022-01-24Opening remarks: The Future of Search and Recommendation
2022-01-24Keynote: Extreme classification for dense retrieval and personalized recommendation
2022-01-24Research talk: Making deep reinforcement learning industrially applicable
2022-01-24Closing remarks: Future of Cloud Networking



Tags:
Search & information retrieval
Microsoft Search
Productivity
search
recommendation
future of search
microsoft research summit