Pengze Liu
 

I am a '21 CS Master student at Dartmouth College. My previous works cover a wide range of Deep Learning Research and Software Engineering. I also have experience in Web and Android development. I am passionate about transforming academic works in engineering practices.

    [news]

  • I will join Arista Networks as a full-time Software Engineer in May 2021.

  GitHub
  LinkedIn
  pengzeliu2 [at] gmail [dot] com


Education

  • Dartmouth College

    Master of Science in Computer Science

    Sep 2019 - Mar 2021 (Expected)

    • Dartmouth CS Scholarship

  • City University of Hong Kong

    Bachelor of Science in Computer Science, First Class Honor

    Aug 2015 - Jul 2019

    • City University Scholarship
    • Chan Sui Hung Best Student Award
    • Dean's List (2016-2019)

Professional Experience

  • Arista Networks

    Software Engineer

    Incoming, May 2021 | Nashua, NH

  • SenseTime Group Ltd

    Research Intern

    Jun 2017 - May 2018 | Hong Kong

    • Accomplished 1K% acceleration while using Python and C++ to upgrade the video preprocessing framework.
    • Improved neural network architecture which resulted in increasing accuracy of pedestrian recognition by 5%.
    • Credited for expanding functions from 7 to 23 on 2.5M samples without precision decay with parameter sharing.
    • Contributed to the internal Caffe-based deep learning framework by converting layers into CUDA C versions.

Publications

  • Localization Guided Learning for Pedestrian Attribute Recognition
    Pengze Liu, Xihui Liu, Junjie Yan, Jing Shao
    British Machine Vision Conference (BMVC), 2018
    Proposed a novel mechanism to bind multi-level features on CNN, achieving state-of-the-art results on 3 benchmarks.

Selected Projects

  • SKINN: Semantic Knowledge Inference Neural Network for COVID-19 Classification
    Python, PyTorch
    A state-of-the-art COVID-19 CT classification network. Accuracy = 95%, AUC = 99%.
    Nov 2020

  • High Performance Conjugate Gradient Solver in CUDA
    CUDA C, C++, Thrust API, OpenMP
    GPU Programming for parallelizable operations. 10-time acceleration.
    Jun 2020

  • Echo: A smart journal App that decorates your quarantine life.
    Android, Computer Vision, NLP, Recommendation System, IBM Watson, Cognitive Science
    Recommend entertainment based on your mood. Check the video below.
    May 2020

  • StayFocused: An Android App to improve working efficiency
    Android, Firebase, SQLite, MVC, Test Driven Development, Git
    A meaningful product to avoid phone addiction.
    Feb 2020

  • MyRuns: Personal Sports Tracking Application
    Android, GPS, SQLite, ML Model Enbedding
    An Android App that tracks your movements and predict the type of sports.
    Jan 2020

  • GrabCut to FCN: Improving Semi-automatic Segmentation with Neural Networks
    PyTorch, OpenCV
    FCN to replace GrabCut. Inspired by the segmentation project shown below.
    Apr 2019

  • BatteryX: Battery Classification in X-ray Images
    Supervised by Dr. Rynson W.H. Lau
    PyTorch, Keras, OpenCV, EdgeBoxes
    Undergraduate thesis. Honestly a trivial work.
    Mar 2019

  • Predicting Suicide Mortality of Schizophrenia Patients
    Tensorflow, Keras, Scikit-learn, Scipy, SVM, Random Forest, DNN, DBSCAN
    4th place in Kaggle competition. Implemented more than 30 ML algorithms/tricks.
    Dec 2018

  • Low-level Segmentation with Deep Learning
    Supervised by Prof. Narendra Ahuja
    MATLAB, PyTorch, OpenCV
    New FCN for low-level segmentation.
    Jul 2018


  • Soccer Player Recommendation System
    Supervised by Prof. Yi Shang
    Hadoop, Apache Spark, MapReduce
    Successfully predicted Alvaro Morata's transfer to Real Madrid.
    Jul 2016

Community Contributions