About Me

Haolin Jia

I will be joining Matthias Grundmann and Tingbo Hou's team in Google AI as a full-time Research Engineer. I will focus on applied research of on-device applications, deep generative models, and real-time animations in Google products, Mediapipe.
I was a master student at New York University MSCS, and received my Bachelor's Degree of Computer Science from Tongji University. In the past, I worked with Prof. Daniele Panozzo on Broad-Phase Collision Detection, and also did research on video/image reconstruction and generation with Prof. Ming-Hsuan Yang.
Here is my complete cv.

News

🎉I am beyond thrilled and humbled to share that I will be joining Matthias Grundmann and Tingbo Hou’s team in Google AI as a full-time Research Engineer! 🚀
🚩I will focus on applied research of on-device applications, deep generative models, and real-time animations in Google products, Mediapipe. 🎨

Experience

  • Google
    Research Engineer, focusing on applied research of on-device applications, real-time animations in Google products, Mediapipe.
    - Mountain View, CA, U.S. -
    Present Jun 2022
  • New York University
    MS in Computer Science
    - New York City, NY, U.S. -
    Jun 2022 Sep 2020
  • University of California, Merced
    Research Intern supervised by Prof. Ming-Hsuan Yang
    - Merced, CA, U.S. -
    Feb 2020 Jul 2019
  • Tongji University
    BA in Computer Science GPA: 91.4/100
    - Shanghai, China -
    Jul 2020 Sep 2016
  • The Experimental High School Attached To Beijing Normal University
    Senior high school in Class One instructed by Jinnan Yuan
    - Beijing, China -
    Jul 2016 Sep 2013

Publication

Semi-Supervised Learning with Meta-Gradient
Xinyu Zhang, Tiahong Xiao, Haolin Jia, Mingming Cheng, Ming-Hsuan Yang,
International Conference on Artificial Intelligence and Statistics(AISTATS), 2021
[Paper, Source code, Poster, Video, Short slides, Full slides]

Project

SmartTab: Sequential Model for Analysing Structure of Articles
Modeled the analysing structures of articles as a Sequence Labeling problem in 2 steps: Paragraph Embedding and Sequence Labeling. Used BERT as a framework of feature extractor to encode paragraphs to gain each sentences embedding. Tagged each paragraph according to its position within the section it belongs to and the relation between it with the neighbor above.
Google, Beijing, China, 2019
[Source code, Poster, Slide]
Real-time system of Steam Market Monitor and Database
Building a database by crawling real-time data from the steam trading market. Designing a data analysis algorithm to predict the tendency of the price based on a structured theoretical financial model mixing with Bayesian estimation and even approaches like a temporal sequential network. Maintaining a front-end website to monitor interested goods and items.
Beijing, China, 2020