2016 INFORMS Annual Meeting Program

100 / 561

INFORMS Nashville – 2016
100
SD19
106B-MCC
Parallel Computing for Optimization and
Data Analysis
Sponsored: Computing
Sponsored Session
Chair: Jonathan Eckstein, Rutgers University, Rutgers University,
Piscataway, NJ, 00000, United States,
1 - The Rectangular Maximum Agreement Problem And Its Data
Analysis Applications
Ai Kagawa, Rutgers University,
The NP-hard rectangular maximum agreement (RMA) problem finds a “box” that
best discriminates between two weighted datasets. Its data analysis applications
include boosting classification methods and boosted regularized regression. We
describe a specialized parallel branch-and-bound method for RMA.
2 - Object-parallel Solution Of Lasso Problems
Gyorgy Matyasfalvi, Rutgers University, 100 Rockafeller Road,
Piscataway, NJ, 08854, United States,
Jonathan Eckstein
We describe an “object-parallel” C++ approach to implementing first-order
optimization methods. As an example application, we solve large-scale Lasso
problems on a distributed-memory supercomputer with the spectral projected
gradient (SPG) method. We can efficiently accommodate highly unbalanced
sparsity patterns.
3 - Asynchronous ADMM-like Optimization Algorithms
Jonathan Eckstein, Rutgers University,
Drawing on some recent work on asynchronous decomposition methods for
monotone inclusions, this talk develops a class of parallel convex optimization
algorithms that resembles the alternating direction method of multipliers
(ADMM) but operates asynchronously. Unlike prior work on asynchronous
variants of the ADMM, the new algorithm’s convergence theory does not rely on
either restrictive assumptions on the problem instance or on random invocation
of subproblems. Instead, it needs only a basic “fairness” restriction that there be
some upper bound on the ratio of the longest and shortest possible subproblem
solution times. Stochastic programming applications may also be discussed.
SD20
106C-MCC
A Unified Framework for Optimization
under Uncertainty
Invited: Tutorial
Invited Session
Chair: Warren B Powell, Princeton University, 230 Sherrerd Hall, Dept
of Operations Research and Financial Eng, Princeton, NJ, 08544,
United States,
1 - A Unified Framework For Optimization under Uncertainty
Warren B Powell, Princeton University, 230 Sherrerd Hall, Dept Of
Operations Research And Financial Eng, Princeton, NJ, 08544,
United States,
Stochastic optimization, also known as optimization under uncertainty, is studied
by over a dozen communities, often (but not always) with different notational
systems and styles, typically motivated by different problem classes (or sometimes
different research questions) which often lead to different algorithmic strategies.
This resulting “jungle of stochastic optimization” has produced a highly
fragmented set of research communities which complicates the sharing of ideas.
This tutorial unifies the modeling of a wide range of problems, from dynamic
programming to stochastic programming to multiarmed bandit problems to
optimal control, in a common mathematical framework that is centered on the
search for policies. We then identify two fundamental strategies for finding
effective policies, which leads to four fundamental classes of policies which span
every field of research in stochastic optimization.
SD21
107A-MCC
Healthcare, General
Contributed Session
Chair: Julie Lynn Hammett, Texas A&M University, 301 Holleman Dr E,
Apt 728, College Station, TX, 77840, United States,
1 - Analysis Of Physician Productivity In Emergency Departments
Krista Foster, University of Pittsburgh, Mervis Hall, Roberto
Clemente Drive, Pittsburgh, PA, 15260, United States,
Jennifer S Shang
We present our analysis of a cohort of U.S. emergency departments. We use visit-
level data to analyze hospital processes and develop models for physician
productivity.
2 - A Review And Extension Of Clinically Significant, Automated
Estimation Of End Systolic And End Diastolic Volumes In
Cardiac MRIs
Michael Kim, Booz Allen Hamilton, 3930 Valley Ridge Drive,
Fairfax, VA, 22033, United States,
We review the winning methods in Kaggle’s Second Annual Data Science Bowl.
The top three algorithms automatically measure endsystolic and enddiastolic
volumes in cardiac MRIs using data from more than 1000 patients. The results
were found to be clinically significant. An analysis of the winning solutions is
presented with a focus on extension through ensembling and transfer learning. In
particular, we architect a machine learning pipeline to extend the top algorithms
to the case of cancer detection given a time series of prostate MRIs.
3 - The Risks Of Risk Adjusted Mortality Rates And A Proposed
Alternative Measure
Thomas Raymond Sexton, Professor, Stony Brook University,
317 Harriman Hall, Stony Brook, NY, 11794-3775, United States,
Christine Pitocco
We consider the widespread use of the risk-adjusted mortality rate (RAMR) to
evaluate hospital performance. We demonstrate that the RAMR, as currently
employed, has significant methodological flaws. We propose an alternative to the
RAMR that is based on standard statistical theory and methods. Applying our
measure yields a more complete and accurate evaluation of hospitals.
4 - Effects Of Artificial Agents Based Ordering On The Supply Chain
Of Perishables
Harshal Lowalekar, Assistant Professor, Indian Institute of
Management-Indore, Prabandh Shikhar, Rau-Pithampur Road,
Indore, 453331, India, ,
Raghu Santanam, Ajay Vinze
We develop a blood bank game which contains a mix of human and computer
based hospital blood banks who order blood at regular intervals from a regional
bank. The objective of all the agents is to minimize their total inventory costs. The
computer agents use a near-optimum policy to determine their order sizes. We
show that presence of a large number of computer based agents in the supply
chain leads to a systematic increase in the order sizes of the hospital banks which
leads to a severe perceived blood shortage in the region. The performance of the
supply chain worsens when the computer agents have the capability to learn
from their past performance.
5 - Remote Patient Monitoring System Framework:
A User Perspective
Julie Lynn Hammett, Texas A&M University, 301 Holleman Dr E,
Apt 728, College Station, TX, 77840, United States,
, Michelle M. Alvarado, Mark Alan Lawley
Healthcare providers are facing an increasing number of patients requiring long-
term care, introducing new challenges to providing fast and affordable care. We
present ongoing research to create a framework for the design, development, and
implementation of remote patient monitoring (RPM) for chronic care. We
highlight the stakeholder needs, system requirements and component
interdependencies. We describe RPM’s need for automated solutions that support
clinical decisions and deliver interventions. This technology must be
interchangeable to suit the varied needs and characteristics of many patients. We
show that these solutions can improve chronic care management.
SD19