Skip to content

Latest commit

 

History

History
131 lines (108 loc) · 6.75 KB

index.md

File metadata and controls

131 lines (108 loc) · 6.75 KB
Raw

Welcome to JackSnowWolf's Blog

Chong, Hu

personal picture

Education

  • Columbia University in the City of New York, NYC, US
    August 2019 - May 2021 (expected)
    • The Fu Foundation School of Engineer and Applied Science
    • Master of Science in Electrical Engineering
  • Shanghai Jiao Tong University (SJTU), Shanghai, CN
    September 2015 - August 2019
    • UM-SJTU Joint Institute
    • Bachelor Degree in ECE; Minor in Data Science

Work Experience

  • MokaHR Company
    December 2018 - April 2019
    Algorithmic Intern, AI Group
    • Combined CTPN and CRNN and developed model to solve resume OCR problems (CN & EN) in TensorFlow and simplified Network Structure and sped up inference time 2s/10s on average, within losing 2% accuracy
    • Adapted open source labeling software to mark text and run evaluation and unit test for different stages
    • Packaged model into web service using gunicorn and Flask, provided API and deployed on Alibaba cloud
    • Implemented cache mechanism with Redis and multistage recognition with high accuracy (over 90% per label) and improved 15% overall performance and about 200% QPS than the original third-party service with parallel processing in Python
  • Beijing Infervision Company
    January 2018 - May 2018 Algorithmic Intern, Modeling Group
    • Used and adjusted YOLOV2 and YOLOV3 under darknet frame and FPN under mxnet for illness detection on DR images
    • Calculated anchor size and number in different methods for YOLO and combined three detection layers to improve accuracy by 5% roughly on tiny objects
    • Utilized Focal Loss to replace original softmax function to care about cases with fewer samples and increase average accuracy by about 3%

Projects

  • HDR (High Dynamical Range) Video Recovering Algorithm
    August 2018 - December 2018
    Company Sponsor: OTC/SSG/Intel; Company Mentor: Zhao Juan, Intel; Academic Advisor: Long Yong
    Report Link

    • Used hdrcnn model to train data to transform LDR to HDR, with different data enhancements and loss functions, eg, cosine loss, in order to reconstructed over exposed area and restored details in dark area
    • Evaluated model performance using HDR-VDP v2 and obtained 20/100 more than traditional mathematical method
    • Applied FFmpeg, OpenEXR to finish the transfer from LDR video to image and image to HDR video, and added meta data of HDR10 format and corresponding BT2020 curves

  • Video Object Segmentation and Visualization in Front-end Using Django August 2019 - December 2019
    Report Link; Proposal Link; Midpoint PPT Link; Final PPT Link

  • Different machine-learning methods to predict popularity of movies using conventional and social media features August 2019 - December 2019
    Report Link; PPT Link

  • Music Recommendation System Analyzed from Million Song Dataset (MSD)
    June 2019 - August 2019
    Academic advisor: Prof. Jing Liu, Prof. Manuel Charlemagne
    Report Link; Poster Link

    • Extracted songs information from 160GB MSD and preprocessed data using hadoop and drill
    • Built similar artist adjacent matrix using MapReduce in hadoop and Spark and compared two methods
    • Used Naive Bayes to guide the scaling data and investigated features inside data using clustering methods
    • Constructed the pipeline of music recommendation based on adjacent matrix and features

  • Bayesian Analysis based on Sample Point Data Using Julia
    June 2019 - August 2019
    Team Member; Academic advisor: Prof. Jing Liu
    PPT Link; Repo Link

    • Plotted graphs of the situation and information using matplotlib to understand data and made basic assumptions
    • Proposed bayesian and mathematical model based on observations and corresponding parameter simulation methods.
    • Implemented Kmean++ clustering in Julia to investigate the location of trees and inferred new data to analysis error

  • Deep Learning Face Super-Resolution with Facial Prior
    May 2019 - August 2019
    Report Link; Proposal Link

    • Re-implemented FSRNet in TensorFlow to reconstruct face high resolution images
    • Dealt with facial landmarks in Helen Dataset and generated and augmented data with different methods
    • Illustrated the significance of using facial prior in face SR tasks with three metrices, MSE, SSIM, PSNR

  • Multi-threaded and Efficient Programming in Database
    Oct 2018 - Dec 2018

    • Implemented table management queries and data manipulation in C++ and handled exceptions in query error
    • Accelerated database using multi-threading and optimizing data structure, enabling 50% faster speed

  • Model Analysis of Effect Factors on Rental Price
    June 2018 - August 2018
    Individual Project; Academic advisor: Prof. Jing Liu

    • Downloaded 958 Mb data from Airbnb and then cleaned, classified and processed data by R
    • Used GAM and GLM to match and explain data; checked model assumptions, outliers, high leverage points and so on to achieve an excellent model effect

  • MCM/ICM: Problem B: Turnpike Toll Plaza Model Based on Queuing Theory February 2017
    Solution Link

    • Built inflow model and merging queue model to generate the incoming cars and calculate the throughput when vehicles exit the toll plazas
    • Designed two new toll plazas: reversible plaza and separate plaza to improve the efficiency
    • Used C++ to realize visualization and MATLAB to generate data and found solutions to different conditions

Computer Skills

  • Programming Language: Python, C++, C, R, Java, MATLAB, Julia, SQL, Verilog.
  • Toolkits/Frameworks: Linux, Hadoop, Drill, Spark, Git, NumPy, pandas, TensorFlow, Matplotlib, OpenCV, LaTeX

Activities

  • Deputy Director, Public Information Department, Student Union of UM-SJTU JI
    July 2016 - July 2017
  • Volunteer Teacher, Elementary School in Yunnan Province
    December 2016 - January 2017
  • Interests: Photography, Football