Bingyuan Liu

Bingyuan Liu

AI researcher and engineer

I bridge AI research and production engineering, with more than a decade of experience across LLM/VLM post-training, multimodal understanding, computer vision, and recommender systems. My work ranges from research published at top venues such as CVPR to AI systems deployed at scale in real-world products.

I am currently an Applied Scientist at Amazon (AWS AI / Amazon AGI), where I work on VLM post-training and multimodal systems for enterprise data. Previously, I was a Research Associate at ÉTS Montréal. Earlier in my career, I held R&D roles at companies such as Coohom and Alibaba, applying AI and machine learning to real-world products and systems.

I received my Ph.D. from the Institute of Automation, Chinese Academy of Sciences, and my B.S. from Zhejiang University.

Education

2010—2015

Institute of Automation, Chinese Academy of Sciences

Ph.D. in Machine Learning

2006—2010

Zhejiang University

B.S. in Information Science and Communication Engineering

Work Experience

Amazon (AWS AI / Amazon AGI)

Applied Scientist

Mar. 2023—PresentVancouver, Canada

  • Led a synthetic data pipeline producing hundreds of millions of examples for document-understanding VLM post-training.
  • Core contributor for VLM post-training, across training tooling, data-mixture design, and evaluation, helping deliver state-of-the-art benchmark results.
  • Developed and shipped multimodal capabilities for Amazon Bedrock Data Automation, including structured information extraction and image and figure summarization.

École de technologie supérieure (ÉTS)

Research Associate

Nov. 2020—Mar. 2023Montreal, Canada

  • Proposed methods for model calibration and image segmentation, resulting in four first-author papers: two at CVPR and two in Medical Image Analysis.
  • Developed eye disease screening solutions with Diagnos.

Coohom, Inc.

Algorithm Expert (Tech Lead, Computer Vision)

May 2018—Mar. 2020Hangzhou, China

  • Led the computer vision team building an image-to-3D furniture retrieval system for KuAiDesign, from model training to production serving.
  • Created and released DeepFurniture: 24K scenes, 170K furniture annotations, and 20K unique 3D identities, with a published benchmark baseline.

Qihoo 360, Co. Ltd.

Machine Learning Engineer

Aug. 2017—Apr. 2018Beijing, China

Developed recommendation systems for Huajiao, a large-scale live-streaming platform.

Alibaba Group, Inc.

Senior Algorithm Engineer

Apr. 2016—Aug. 2017Hangzhou, China

Built production OCR systems for business documents and computer-vision components for Alibaba Cloud PAI.

Huawei Technologies, Co. Ltd.

Software Engineer

Jul. 2015—Apr. 2016Nanjing, China

Developed applied machine-learning solutions for enterprise partners, including China Mobile.

Selected publications

All publications
  1. Reducing Distraction in Long-Context Language Models by Focused Learning

    arXiv, 2024 · Second author

  2. Class Adaptive Network Calibration

    CVPR, 2023 · First author

  3. The Devil is in the Margin: Margin-based Label Smoothing for Network Calibration

    CVPR, 2022 · First author

  4. Furnishing Your Room by What You See: An End-to-End Furniture Set Retrieval Framework with Rich Annotated Benchmark Dataset

    arXiv, 2019 · First author

Contact

Email is the best way to reach me: liubingyuan1988@gmail.com