Sam Voisin

Hillsborough, NC

Welcome to my website! My name is Sam Voisin. I am a data science, machine learning and software design professional. I have extensive experience in natural language processing (NLP), computer vision (CV), time series analysis and Bayesian methods.

🦾 I am passionate about intuitive, gesture-based control systems. I am currently developing GestuReMote, a python library which provides an interface for controlling your desktop/laptop with physical gestures.

🧙‍♂️ Another project I am working on is AIDM, an LLM-powered chatbot that can act as a virtual “Game Master” for tabletop role-playing games. AIDM utilizes a combination of prompt engineering and software engineering features to ensure the LLM produces appropriate text.

⛵️ I currently work as a Senior Data Scientist at Tradewind Data Science where I build value for clients by developing inferential and predictive models to guide strategy and policy decisions.

I graduated from Duke University with a Master’s in Statistical Science in 2020. While at Duke, I was exposed to a broad range of fields that are important to data science. In addition to probability and statistics courses provided by the Department of Statistical Science, I took courses in algorithms, advanced linear algebra, topology, high-dimensional data analysis and topological data analysis (TDA). My master’s thesis Graph Diffusion for Gesture Classification focused on developing techniques to improve the accuracy of systems that classify physical gestures.

When I’m not thinking about computers I enjoy spending time with my family, reading, learning new skills, and traveling!

latest posts

Dec 1, 2024	Lessons of Vertical Slice Architecture
Jan 15, 2024	R-Squared and Non-Linear Models
Dec 23, 2023	The Coefficient of Determination and Adjusted R-Squared

selected publications

2023

Topological Simplification of Signals for Inference and Approximate Reconstruction

Gary Koplik, Nathan Borggren, Sam Voisin, and 4 more authors

In 2023 IEEE Aerospace Conference, Mar 2023

Abs

As Internet of Things (loT) devices become both cheaper and more powerful, researchers are increasingly finding solutions to their scientific curiosities both financially and com- putationally feasible. When operating with restricted power or communications budgets, however, devices can only send highly- compressed data. Such circumstances are common for devices placed away from electric grids that can only communicate via satellite, a situation particularly plausible for environmental sensor networks. These restrictions can be further complicated by potential variability in the communications budget, for ex-ample a solar-powered device needing to expend less energy when transmitting data on a cloudy day. We propose a novel, topology-based, lossy compression method well-equipped for these restrictive yet variable circumstances. This technique, Topological Signal Compression, allows sending compressed sig-nals that utilize the entirety of a variable communications budget. To demonstrate our algorithm’s capabilities, we per-form entropy calculations as well as a classification exercise on increasingly topologically simplified signals from the Free- Spoken Digit Dataset and explore the stability of the resulting performance against common baselines.

2022

Topological Feature Tracking for Submesoscale Eddies

Sam Voisin, Jay Hineman, James B. Polly, and 6 more authors

Geophysical Research Letters, Mar 2022

e2022GL099416 2022GL099416

Abs

Abstract Current state-of-the art procedures for studying modeled submesoscale oceanographic features have made a strong assumption of independence between features identified at different times. Therefore, all submesoscale eddies identified in a time series were studied in aggregate. Statistics from these methods are illuminating but oversample identified features and cannot determine the lifetime evolution of the transient submesoscale processes. To this end, the authors apply the Topological Feature Tracking (TFT) algorithm to the problem of identifying and tracking submesoscale eddies over time. TFT identifies critical points on a set of time-ordered scalar fields and associates those points between consecutive timesteps. The procedure yields tracklets which represent spatio-temporal displacement of eddies. In this way we study the time-dependent behavior of submesoscale eddies, which are generated by a 1-km resolution submesoscale-permitting model. We summarize the submesoscale eddy data set produced by TFT, which yields unique, time-varying statistics.