Data Science

Data Science Interests

Research-wise, I'm interested in the interpretability of RNNs. I have completed coursework in deep learning, classical machine learning, and computer vision.

* * *

Experience:

Researcher May 2020 → Sep. 2021

University of Maryland, College Park (Part-Time)

Investigated deep learning models for time-series under Professors Soheil Feizi and Hector Bravo (now of Genentech), and PhD student Aya Ismail.
Trained neural networks (DeepAR, LSTMs, Transformers, TCNs) on university servers with SLURM, and presented results at meetings. Written in Python with PyTorch.
Extended recent work providing provable guarantees on neural network convergence by exploring the data augmentation of time-series via spherical harmonics expansion.

Personal Projects:

PyTorch Implementation of DeepAR Jan. 2021

Github

Implemented Amazon Research’s DeepAR neural network in PyTorch using their 2020 paper.
Trained this autoencoder for regression tasks on time-series.

Synthetic Time-series Generator Jun. 2020

Github

Generated 75,000 synthetic datasets in CSV format for an ongoing research paper on recurrent neural networks
Wrote framework supporting the creation of AR, ARMA, ARMAX, and non-linear stochastic processes
Written in C++17, with Catch2, and Boost Math and Multiprecision

Data Science Courses:

CMSC422

Introduction to Machine Learning

CMSC426

Computer Vision

CMSC498L

Introduction to Deep Learning