MODEL // SIMULATE // LEAD // ENDURE // FIRST LIGHT

Trevor Caskey.

I build models of complex systems, chase East Coast sunrises, and train for the long haul. Plays them loud. Runs them long.

◆ Ph.D. Systems Eng.◇ Adjunct Instructor● Ultra runner■ Hard rock & reggae◒ Sunrise chaser
01 — Pre-race briefing

Long horizons.
Complex systems.

I love to develop and formulate models of complex systems — it's my passion. I enjoy the full pipeline of the analytic process: data collection, storage and manipulation, model development and execution, and visualizing results so decision-makers can see what the model sees.

Today I lead projects and engineering teams, and teach as adjunct faculty. Ultra running and modeling share a mindset: break a huge problem into segments, plan for uncertainty, and keep moving when the variance shows up. Hard rock keeps the cadence up.

0yrs
Modeling in the field
0
Degrees · math → OR → SE
0yrs
Teaching · 4 universities
50mi
Longest run
LIVE MODELCircle-pit order 0.00
Agent-based mosh pit · ● moshers self-propel & flock, ● crowd just gets pushed. Flocking keeps the circle pit spinning; drop flocking or crank the noise and it breaks into a mosh pit. Switch to Reggae sway for the summer-set version: everyone rides one slow, shared rhythm instead of a pit. Click the crowd to recruit more moshers.
After Silverberg, Bierbaum, Sethna & Cohen, PRL 2013 (moshpits.js).
02 — Course profile

The career, as an elevation chart.

Start · 2007IndustryTeachingNow
03 — Race bibs

Credentials.

George Washington
University
2018

Ph.D. Systems Engineering

George Washington University

George Mason
University
2011

M.S. Operations Research

George Mason University

William & Mary
2007

B.S. Applied Mathematics

College of William & Mary

04 — Cross-training

Off the model.

▲ TRAIL

Ultra running

Long days on dirt. Pacing, fueling, and failure-mode analysis at mile 40 — a 50-miler is just a stochastic process with aid stations.

?? miraces coming sooncity, st · [hh:mm:ss]
● ON REPEAT

Two sides

Side A for long miles and model runs, distortion on. Side B for summer: warm reggae-rock, the beach, and first light.

A1songs coming soonROCK
05 — Field research

Published models, playable.

▲ Agent-based · game theory · belief learning

Deter and protect: Hot Spot Patrol

Caskey, Wasek & Franz (2018). Deter and protect: crime modeling with multi-agent learning. Complex & Intelligent Systems 4:155–169. doi:10.1007/s40747-017-0062-8 · open access
DAY 0 · click a block to send a patrol
targetsoffendersguardianscrime
0crimes
—patrolled block, during vs before
Patrol intensity (share of guardian time)

Routine activity theory: crime needs a motivated offender and a target in the same place, with no capable guardian. Every agent here learns where the others tend to be (weighted fictitious play over 14 days) and picks a block by softmax. Send a patrol and crime falls in that block and spreads to its neighbours; when the patrol ends, the guardians' habits fade and crime drifts back, the "inverse displacement" the paper fit to Seattle, Austin and Minneapolis hot-spot data. Simplified browser re-implementation of the paper's Eqs. 4–12 (5 blocks, 50 targets, 10 offenders, 5 guardians, h = 14, αO=.5, αT=αG=1). Illustrative, not calibrated to case-study data.

06 — Drop bag

The kit.

◷
Watch

Garmin Forerunner 970

GPS, HR, and the data feed for every post-race model.

Feeds
GPS · HR
▲
Shoes

Saucony Ride 19

Cushion and comfort for long miles.

Mount to Coast H1 & Teva R2T

Grip for the transition from road to technical trails.

Road
Ride 19
Road → trail
H1 · R2T
≋
Socks

Injinji Trail Midweight

Toe socks. Blister prevention is a reliability problem.

Type
Toe socks
⚡
Fuel

Maurten + SiS + Scratch Labs

Water, carbs, calories, sodium, and protein on a schedule — fueling is a control loop.

Inputs
H₂O · Carbs · kcal
Sodium · Protein
07 — Finish line

Let's go the distance.

Email me LinkedIn