Kevin Urban

About Kevin Urban

I develop statistical and machine learning approaches that transform noisy physiological sensor data into reliable and clinically meaningful metrics and risk indicators for remote patient monitoring, early warning alerts, and clinical decision support.In many projects I contribute most strongly in ambiguous or early-stage problem spaces, where framing the right question is often more valuable than optimizing the wrong model. I\'m particularly effective at clarifying poorly defined technical problems, stress testing assumptions, and identifying structural risks early on.Much of my career has focused on extracting signal from complex dynamical systems observed through noisy, partial, and indirect measurements. My work sits at the intersection of statistical modeling, time series machine learning, and real-world sensing environments. I focus on the challenges that arise when models move beyond curated research datasets into messy operational settings: irregular sampling, noisy signals, evolving baselines, and uncertain ground truth. This work has included simulation frameworks, adaptive statistical estimators, Bayesian statistics, and deep neural networks applied across sensing modalities, from wearable motion sensors and physiological biosignals to image and video data.Themes that run through my work include:• Extracting signal from noisy or partially observed systems• Designing statistical models and monitoring approaches that remain robust under real-world conditions• Bridging research models with the realities of operational constraints• Translating between research, engineering, clinical, and product perspectivesI\'ve applied statistical and machine learning methods to diverse physical and physiological phenomena such as:• Solar wind prediction and space physics modeling using distributed sensor networks across ground-based, airborne, and space-based instruments• Computational modeling of nonlinear dynamics and granular fluids• Deep learning and signal processing for wearable biosensors and neurological disorder detection (Parkinson’s, Rett syndrome, sleep apnea)• Computer vision tasks such as segmentation and analysis of thermograms, images, and video• Pattern recognition and anomaly detection in high-dimensional multimodal time series data• Real-time analytics infrastructure and dashboards for large-scale digital media platforms (WWE)My background includes a PhD in Physics and research experience at NASA, where I first developed an interest in complex systems and inferring hidden structure from noisy data.

Experience

1. Principal Data Scientist

   Podimetrics

   Dec 2024 — Present

   Developed statistical monitoring and modeling systems that transform noisy physiological sensor data into clinically actionable signals for diabetic foot ulcer (DFU) and fall risk detection.* Built a simulation and tournament framework to benchmark online statistical models for patient stability monitoring and fall-risk assessment across synthetic scenarios (drift, regime shifts, outliers, irregular sampling) using multi-objective scoring (prediction error, probabilistic likelihood, and alert quality), as well as real-world patient data for visual model validation.* Developed adaptive streaming baseline estimators that transition from population priors to personalized patient models using a NIG Bayesian cold start, homotopy-style handoffs, and recursive filtering with time-aware decay.* Advanced modeling roadmap toward Kalman-style filtering separating process and measurement noise and addressing long observation gaps, improving reliability and interpretability of clinical alerts.* Developed a synthetic thermogram generation framework to simulate DFU development, enabling controlled experiments for evaluating detection algorithms and validating improvements to asymmetry-based risk scoring.* Led data validation and model adaptation during the transition from SmartMat to SmartMat+, identifying experimental design flaws and dataset integrity issues that affected thermal curve modeling, and collaborating with R&D to improve data quality and evaluation procedures.* Evaluated and improved thermogram extrapolation and calibration models used to predict equilibrium temperatures from short scans, developing physics-informed alternatives based on Newtonian heat transfer to benchmark and potentially replace existing kNN approaches.* Investigated and resolved data pipeline and sensor-processing issues by analyzing raw scans, improving calibration models, and collaborating with software engineers to diagnose edge cases affecting clinical signal quality.

Education

• New Jersey Institute of Technology

  Bachelor's degree, Applied Physics (Minor: Applied Math)

  2003 — 2008

• Udacity

  Nanodegree, Deep Learning

  2017 — 2017

• New Jersey Institute of Technology

  Master's degree, Applied Physics (Minor: Applied Math)

  2008 — 2010

• New Jersey Institute of Technology

  PhD, Physics

  2012 — 2016