About Daniel Jordon

I am a Machine Learning Engineer at WeWork. Previously, I was a Data Scientist at SeatGeek and before that I was a post-doctoral researcher working with Dr. Robert Hampshire at the University of Michigan’s Transportation Research Institute and at the Mobility Analytics Thrust at Heinz College at Carnegie Mellon University. My work focuses on statistical learning applied to intelligent transportation, specifically applying reinforcement learning tools to parking and routing using vehicle-2-vehicle communication technology (no one would give me funding to study unintelligent transportation). I used queueing theory and simulations to evaluate learning algorithms that can aid in routing drivers to their destinations while taking parking availability into consideration. I did my PhD in Mathematics at Drexel University, studying dynamical systems, and PDEs.

Projects

Polygon-partitioner (Github repo). A collection of methods related to polygon approximation and polygon partitioning. There is a method that can partition an arbitrary orthogonal polygon into a set of non-overlapping rectangles. It also implements algorithms that allow you to cover a polygon by an orthogonal polygon with arbitrary precision.

Queueing-tool (Github repo, pypi, documentation). Queueing-tool is an agent based simulation framework for analyzing queueing networks. The visualizations are managed using matplotlib while the agents and queueing interactions are handled by the package. This package serves as the basis for my reinforcement learning and approximate dynamic programming research.

Preprints

  • Properties of differential operators with vanishing coefficients. Submitted (2015). (arXiv:1504.03157)

Publications in preparation

  • Dynamic routing and parking: Using reinforcement learning and queueing theory for efficient routing and parking, (with R. Hampshire). We develop an algorithm whereby a driver’s navigation unit can learn how to efficiently route the driver to their destination and quickly park. The algorithm relies on drivers sharing a summary statistic of their ‘experiences’ when finding parking. The summary that is shared is opaque enough to ensure that a person’s driving habits are not recoverable from the summary statistic.

  • Estimating parking occupancy and cruising using meter payment data, (with R. Hampshire). We develop a method that estimates how many people are currently parked and how many people are looking for parking given parking payment data. We then test our method against a large parking dataset from SF park.

    I gave a talk about this work at CAARMS 2016. The conference was held at Princeton University and the Institute for Advance Study (IAS). I gave my talk at the IAS, so the talk was recorded (and I didn’t find out about that fact until after my talk).