Sun Woo

Sun Woo Kim


Indiana University Bloomington

PhD, Computer Science — 3.8 GPA

University of Illinois at Urbana-Champaign

BS, Physics — 3.3 GPA

Research Experience

School of Informatics, Computing, and Engineering

Bloomington, IN

Independent Study

8/2017 - Present
  • Perform source separation on quantized input and Gated Recurrent Unit (GRU) cells by learning on floating points then transferring over to a binarized version of the model

National Center for Supercomputing Applications

Urbana, IL

CyberGIS SPIN Intern

5/2015 - 5/2016
  • Designed and developed a parallel program employed in CyberGIS Summer School on Big Data Landscapes
  • Created a prototype for an interactive web application with ExtJS that displayed Twitter activity within major cities in the United States along with plots drawn with D3.js


Binary GRU

1/2018 - Present

  • Mix clean signals with noises to generate input spectrums and quantize with Lloyd Max Quantization
  • Create a custom GRU Cell in Tensorflow using binarized weights and activations with custom gradients
  • Build a $n$-layered RNN model to save and restore trainable variables
  • Quantize weights by binarizing, introducing sparsity, and scaling using external trainable variables
  • Visualize changes in weights and respective gradients using Tensorboard

Identification and Localization of Siren Signals

3/2017 - 5/2017

  • Constructed a siren detecting architecture distinct from existing work using a fast singular vector model classification method with sklearn, specifically non-negative matrix factorization (NMF) and support vector machines (SVM)
  • Trained a dimensionality reduction model using limited training data and performed localization with the returned set of general basis vectors
  • Experimented on simulated data and showed accurate estimation of ambulance location

Multitask Learning Accura-SEA on Stock Prices

10/2016 - 12/2016

  • Attempted to use historical stock price data to predict future directions through Multitask Learning
  • Studied the relationship between the number of tasks and the accuracy performance of Multitask Factorized Gradient Descent and other regressional algorithms
  • Evaluated the performance of each models through bootstrapping and meta-parameter selection using internal k-fold cross-validation

Champaign County Bikes CIGI Hack-a-Thon

12/2015 - 1/2016

  • Utilized dojo/request package to provide asynchronous requests to OSRM's viaroute API for retrieving the shortest path between given coordinates
  • Drew the route as a polyline with flags for beginning, middle, and end points using ARCGIS API's Graphic class

Urbanflow Web Application

12/2015 - 1/2016

  • Collaborated with 2 post-docs to create a prototype for a web application to show the movement of users within cities through analysis on Twitter data
  • Developed with ExtJS to create diverse panels with interactive grids for each user input category that responded by displaying plots drawn with D3.js or zooming into the extent of selected vector files using OpenLayers 2 library

Jenkins Plugin Extension

9/2015 - 12/2015

  • Worked in a team of 7 in a test driven development to extend an existing Jenkins Plugin, Test Results Analyzer, following XP methodology
  • Added more functionality such as comparing build results by selecting the checkboxes above build numbers and easing user navigation by displaying links to the results page

Eye Tracker Analysis

6/2015 - 12/2015

  • Analyze .txt data files from Tobii software on 39 test subjects for eye position, duration, and response and create visuals using D3.js and Bokeh
  • Build interactive website for other researchers to explore the data and plots using Javascript/jQuery, HTML, CSS, PHP, and MySQL

Parallel Terrain Analysis and Predictive Mapping

5/2015 - 7/2015

  • Partitioned and distributed rows of the input image to number of processes desired by the user and calculated slope gradients in parallel through the Message Passing Interface (MPI) on HPC
  • Performed k-means clustering to create a terrain classifier with Apache Spark's MLlib

Teaching Experience

Indiana University Bloomington

Bloomington, IN

Associate Instructor for E533: Deep Learning Systems

1/2018 - 5/2018
  • Answer questions on deep learning concepts, implementations with Tensorflow, and setting up with GPU on servers during office hours
  • Assist guest lecturers in preparing, taking attendance, and recording the lectures

Technical Skills


Python, Tensorflow


C/C++, Java


LaTeX, HTML, Javascript, MySQL

Sun Woo Kim — — (434) - 987-9753