Qingyun Liu

About Qingyun Liu

I focus on AI for both industrial-scale production and research.Currently, I\'m a Research Engineer in DeepMind, working on Generative AI for personalization. Specifically, I focus on aligning Gemini with user behaviors to develop unified recommendations at Google scale like YouTube products.Previously, I was in Google Brain working on neural network modeling specialized in recommendation for Google products, e.g. multi-task learning, neural deep retrieval, user modeling.Before that, I was a Research Scientist at Facebook and fought against various integrity challenges in both News Feed and Groups, e.g. anti- abuse signal development/measurement/remediation for user behaviors/complex entities, build ML- based detection systems. I also worked on understanding fundamental problems of online communities, e.g. community leadership persona, well-being of community leaders.I finished my Ph.D. in Computer Science at University of California, Santa Barbara. My research interests are centered around large-scale data analysis, with a focus on online social networks (OSNs). Most of my work is measurement-driven and modeling based, using data mining and machine learning techniques. Examples include graph modeling (e.g. link prediction, graph isomorphism), human online behaviors studies (e.g. user temporal behavior modeling, quality and social structure of online forums). I also have rich internship experiences in both academia (MSR, Technicolor Research, AT & T Research) and industry (Facebook), with projects on ranking, advertising, network traffic anomaly monitoring, etc.

Experience

1. Research Engineer

   Google DeepMind

   Nov 2019 — Present

   Conduct AI research and push the frontier of modeling for Google personalization, e.g. YouTube- April 2023 to present: Google DeepMind, focus on Generative AI- Nov. 2019 to April 2023: Google Brain (in Google Research), focused on large-scale neural modeling in ranking/nomination/user modeling, e.g. multi-task learningSelected publications- LEVI: Generalizable Fine-tuning via Layer-wise Ensemble of Different Views (ICML 2024)- Co-optimize Content Generation and Consumption in a Large Scale Video Recommendation System (RecSys 2024)- Short-form Video Needs Long-term Interests: An Industrial Solution for Serving Large User Sequence Models (RecSys 2024)- Multitask Ranking System for Immersive Feed and No More Clicks: A Case Study of Short-Form Video Recommendation (CIKM 2023)- Talking Models: Distill Pre-trained Knowledge to Downstream Models via Interactive Communication (arXiv 2023)

Education

• Peking University

  Bachelor of Science - BS, Intelligence Science and Technology (School of Electrical Engineering and Computer Science)

  2008 — 2012

• UC Santa Barbara

  Master of Science (M.S.), Computer Science

  2012 — 2016

• UC Santa Barbara

  Doctor of Philosophy (Ph.D.), Computer Science

  2012 — 2017