Roman Z. Wang Machine Learning Engineer. Making sense of the world through machine learning.

About

Hi, I’m Roman.

I’m a part-time Computer Scientist at Johns Hopkins Applied Physics Laboratory and part-time Master's student in CS at Columbia University. I’m curious and always looking to take on new challenges. My passion is using machine learning and artificial intelligence to better understand the world around us and engineer solutions to previously unsolvable problems.

When I'm not working at the screen, I’m learning jiu-jitsu and judo, or doing photography.

Projects

Recognizing Families in the Wild 2021 Challenge

Competition winner for RFIW 2021 by placing top-3 in the competition with team. We engineered an ensembles of human-engineered models and programs used by program synthesis.

Presented at FG 2021. GitHub

Vehicle Routing Trifecta Research Paper

Multi-objective vehicle routing system tested on open source data.

Presented at ICCCN 2019.

Sepsis Research Paper

Developed models for papers using the MIMIC-III database. Tons of data cleaning was involved and weakening assumptions about the data was a must. Database extraction scripts are included in the database. Presented at ICHI 2018.

Presented at ICHI 2018. GitHub

Multi Agent Tanks Reinforcement Learning Challenge

Designed a proprietary PPO-LSTM reinforcement learning model that learned through tournament self-play and engineered video features for a workplace AI challenge.

Simulation Environment.

Polynomial Expansion with Machine Learning

Used a transformers to symbolically expand single-variable second-ordered polynomials. I deployed model as a REST API via Google App Engine. Use python math syntax e.g. c*(c-2) or 2*c**2, not (9-x)(8+x) or x*(9x-1). Performance is still being improved.

Endpoint docs.

GPU-enabled NTRUEncrypt Encryption

Implemented an NTRUEncrypt encryption algorithm, an NTRU Lattice-Based encryption algorithm. As of January 1st, 2022, NTRUEncrypt had made it to the most recent round of the NIST Post-Quantum Cryptography Standardization competition.

GitHub

Delaunay Triangulation Comparison

Implemented the incremental-insertion Delaunay Triangulation algorithm found in Primitives for the Manipulation of General Subdivisions and the Computation of Voronoi Diagrams by Guibas and Stolfi; and a divide-and-conquer algorithm found in A faster divide-and-conquer algorithm for constructing delaunay triangulations by Dwyer. This project idea was the final project of CS274 from Berkeley during the spring 2019 semester.

Project Idea. GitHub

Ray Tracing

Implemented a ray tracer from scratch.

GitHub

Multi-layer Perception

Implemented a multi-layer perception from scratch.

GitHub

Skills

Bio

September 2021 -
Graduate Research Assistant
@ Columbia University Department of Computer Science
September 2019 -
Computer Scientist
@ Johns Hopkins Applied Physics Laboratory
September 2016 - September 2018
Undergraduate Research Assistant
@ University of Virginia Department of Systems and Information Engineering
June 2018 - September 2018
Software Engineering Intern
@ Northrop Grumman
July 2017 - September 2017
Undergraduate Research Assistant
@ University of Virginia Department of Computer Science
June 2014 - August 2014
Information Technology Intern
@ Henrico County Government

Education

2019 - Columbia University
M.S. Computer Science
Natural Language Processing
2015 - 2019 University of Virginia
B.A. Computer Science
B.A. Mathematics
2011 - 2015 Deep Run High School
Center for Information Technology
Valedictorian
Presidential Scholar Candidate