Matthew Shumway

I'm a PhD student at Boston University (BU) Faculty of Computing and Data Sciences (CDS). I am fortunate to work with Xuezhou Zhang, focusing on reinforcement learning.

Prior to BU, I obtained my bachelors at Brigham Young University (BYU), studying Applied and Computational Mathematics. During my undergraduate, I was fortunate to work with Mark Kempton on studying non-backtracking random walks.

Email / CV / LinkedIn / Github

Publications

On defining Kemeny's constant for non-backtracking random walks
Jane Breen, Mark Kempton, Adam Knudson, Matthew Shumway
Preprint, 2025
arXiv / code

We propose two different possible definitions of defining Kemeny's constant of a graph based on non-backtracking random walks.

Work Experience

	Vivint AI Engineer Intern May - August 2025 Explored the capabilities of VLMs for the curation of personalized user highlight videos from home security camera footage.
	bioMérieux R&D Data Science Intern July - September 2024 Developed traditional ML models and collected data to predict the impact of DNA degeneracy on PCR-based disease detection assays.

Fun Projects From Undergrad

Selected | All

	Optimal Control Approach to Wildfire Rescue Max Fennimore, Josh Hoagland, Eli Sampson, Matthew Shumway BYU MATH 438: Modeling Dynamics and Control II, 2025 pdf / code We use Pontryagin's Maximum Principle to solve a boundary value problem to determine rescue paths that minimize time, risk, and control input.
	Introductory Exploration to Contextual Bandits Matthew Shumway BYU CS 677: Bayesian Methods for CS, 2025 pdf Explores some classical and modern algorithms for contextual bandits, including LinUCB, Thompson Sampling, and neural network based methods.
	The Bulls and the Bears: Regime Identification and Forecasting Michael King, Trevor Larsen, Matthew Shumway, Nathan Todd, Brandon Waits BYU MATH 404: Modeling with Uncertainty and Data II, 2025 pdf Uses clustering algorithms, hidden Markov models, and a statistical jump model to classify historical market regimes and forecast future predictions on which an investment strategy is built.
	AlphaGomoku: Applying AlphaZero to the Game of Gomoku Matthew Shumway, BYU CS 401R: Dynamic Programming and Reinforcement Learning, 2024 pdf / code Uses a deep reinforcement learning approach based on AlphaZero to learn the game of Gomoku, demonstrating the effectiveness of self-play and neural network-guided search in a simplified 6x6 board setting.
	Predicting Pickup Density of Taxi Cabs in NYC Cole Edgren, Eli Sampson, Matthew Shumway BYU MATH 402: Modeling with Uncertainty and Data I, 2024 pdf / code Using a LightGBM model to predict taxi cab pickup density in New York City based on historical data and temporal features.
	Alternatives to Backpropagation: Exploring the Forward Forward Algorithm Max Fennimore, Jason Oliphant, Eli Sampson, Matthew Shumway, BYU MATH 522: Mathematical Fundamentals of Deep Learning, 2024 pdf / code An exploration of Geoffrey Hinton's Forward Forward algorithm as an alternative to backpropagation, including implementation details and performance comparisons on standard datasets.
	Analysis of Market Competition Using ODE Models Cole Edgren, Josh Hoagland, Matthew Shumway, Nathan Todd, Brandon Waits BYU MATH 436: Model Dynamics and Control I, 2024 pdf / code Uses a system of ODEs to model the competition betwen companies for market share.

Honors and Rewards

CDS Duan Family Fellowship, 2025
Edwin S. Hinckley Scholarship, 2024

Template based on Jon Barron's site.