[FRIAM] Fwd: Thesis Proposal - Dylan Fitzpatrick - Today - Monday, May 4 at 9am - via Zoom

[George Duncan]

Most appropriate topic

George Duncan
Emeritus Professor of Statistics, Carnegie Mellon University
georgeduncanart.com
See posts on Facebook, Twitter, and Instagram
Land: (505) 983-6895
Mobile: (505) 469-4671

My art theme: Dynamic exposition of the tension between matrix order and
luminous chaos.

"Attempt what is not certain. Certainty may or may not come later. It may
then be a valuable delusion."
>From "Notes to myself on beginning a painting" by Richard Diebenkorn.

"It's that knife-edge of uncertainty where we come alive to our truest
power." Joanna Macy.




---------- Forwarded message ---------
From: Michelle E Wirtz <mwirtz at andrew.cmu.edu>
Date: Mon, May 4, 2020 at 6:29 AM
Subject: Thesis Proposal - Dylan Fitzpatrick - Today - Monday, May 4 at 9am
- via Zoom
To: heinz-faculty at lists.andrew.cmu.edu <heinz-faculty at lists.andrew.cmu.edu>,
heinz-phd at lists.andrew.cmu.edu <heinz-phd at lists.andrew.cmu.edu>, Daniel
Neill <daniel.neill at nyu.edu>, Roni Rosenfeld <Roni.Rosenfeld at cs.cmu.edu>
Cc: Diane L Stidle <stidle at andrew.cmu.edu>


*Friendly reminder – *



Hi all,

Please join us today, Monday, May 4, 2020 via Zoom at 9am when Dylan
Fitzpatrick will be presenting his thesis proposal.

*Title:* Predicting Health and Safety: Essays in Machine Learning for
Decision Support in the Public Sector

*Thesis committee: *Daniel Neill, Rayid Ghani, Wilpen Gorr, Roni Rosenfeld



*Zoom Link:*

https://cmu.zoom.us/j/95758239810?pwd=RmhFL1hDY3pYUzJTWC9GMzBCdndnUT09

*Meeting ID:* 957 5823 9810
*Password:* 032643

*Abstract:  *Public service agencies are increasingly turning to machine
learning techniques for support in settings where accurate predictions or
characterization of patterns in spatiotemporal data can improve
social conditions. This thesis presents three case studies in which we
propose novel methods to inform operational decisions in the domains of
public health and safety.

First, we present a subset scan approach for detecting localized and
irregularly shaped anomalous patterns in spatial data. The proposed method
iterates between a penalized fast subset scan and a kernel support vector
machine classifier to accurately detect spatial clusters without imposing
hard constraints on the shape or size of the anomalous pattern. We
demonstrate the performance of this approach in simulated experiments and
on the real-world applications of disease outbreak detection, crime
hot-spot detection, and pothole cluster detection.

Second, we leverage prescription drug monitoring data to assess risk of
opioid misuse based on individual-level opioid timelines. We introduce a
shape-based clustering framework to evaluate risk of misuse in new
individuals when patient outcomes are unknown. We also develop a new method
for semi-supervised learning with recurrent generative adversarial
networks, designed to assess risk of opioid misuse in new patients when
labeled instances of unsafe drug use are available but sparse.

Last, we discuss the design, implementation, and evaluation of a
hot-spot-based predictive policing program in Pittsburgh, PA, highlighting
results from a randomized field trial. We find statistically and
practically significant reductions in violent crime counts within treated
hot spots, and find minimal evidence of crime displacement to other areas
resulting from increased patrols to treated areas.



*Link to paper: *
https://www.dropbox.com/s/h6l151fs7k8uzf5/Fitzpatrick_proposal.pdf?dl=0
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20200504/c0cc8b15/attachment.html>