Integer Programming Techniques for Matroid Circuit Problems
John Arellano
Computational and Applied Mathematics
Rice University
Although some combinatorial optimization problems associated with matroids can be solved in polynomial time, finding particular circuits in matroids is an NP-hard problem. It is related to compressive sensing and finding the degree of redundancy of sensor networks. We attempt to solve these types of problems to optimality using integer programming techniques and present computational results.
A Computational Study of Decomposition Algorithms for Stochastic Programs with Mean-Risk Objectives
Tanisha G. Cotton
Industrial and Systems Engineering
Texas A&M University
To introduce risk into linear stochastic programs, convexity preserving dispersion statistics, quantile- deviation and absolute semideviation can be used to represent mean-risk objectives. In this poster presentation, we report on a computational study of stage-wise decomposition algorithms for this class of stochastic programs.
Fast Generalized Subset Scan for Anomalous Pattern Detection
Edward McFowland III
Heinz School of Public Policy and Management
Carnegie-Mellon University
We propose Fast Generalized Subset Scan (FGSS), a new method for detecting anomalous patterns in datasets with categorical variables. We frame the pattern detection problem as a search over subsets of data records and attributes, and exploit a novel property of the nonparametric scan statistic that allows for efficient optimization over subsets without an exhaustive search over the exponentially many subsets. As a result of this efficient optimization, we can quickly find the subset of records that is optimal for a given set of attributes; similarly, we can efficiently optimize over all subsets of attributes for a given subset of records. The algorithm iterates between maximizing over records and attributes until it converges to a local maximum, which represents a group of anomalous records and the set of attributes for which they are anomalous. Choosing the maximum of multiple randomly restarted searches discovers the global maximum with high probability.
Our results demonstrate that FGSS can successfully detect useful anomalous patterns in various application domains, including disease surveillance, customs monitoring, and network intrusion detection. FGSS dramatically reduced run-times and achieved higher detection power than current methods on massive multivariate dataset.
Risk-Based Technology Assessment for Capital Equipment Decisions in Small Firms
Samuel Merriweather
Industrial and Systems Engineering
Texas A&M University
Within this presentation we discuss a risk-based approach to capital equipment budgeting and acquisition of new equipment and/or technologies in small firms. The approach overcomes deficiencies in the capital budgeting process by connecting equipment use to projected cash flows via discrete-event simulation. A healthcare application is illustrated.
Transient Analysis of the Border Crossing Process Using Congestion Based Policies
Hiram Moya
Industrial and Systems Engineering
Texas A&M University
Trade is the U.S. depends on an efficient flow of inspected containers in and out of the border ports of entry (POE), while focusing on security, and being cost effective. This research focuses on all commercial traffic at a southern border POE, where there is a non-steady state, terminating system. Using transient analysis, we present analytical and experimental results of congestion based policies with a fixed number of servers, by implementing a primary inspection station service switch.
Differential Equations Modeling of Patients and Physicians Dynamics in Emergency Rooms: Issues and Optimal Control Staffing Policies
Jerome Ndayishimiye
Industrial and Systems Engineering
University of Buffalo
Hospital emergency rooms are difficult to manage because of the complexity of allocating costly resources, mainly physicians, in the light of the dynamical arrivals of patients and the costs of delayed medical care. We propose using ordinary differential equations to model the dynamics of patients and physicians and then use optimal control theory to determine optimal physicians allocation policy. For a practical implementation of the policy, we use a heuristic algorithm based on the concepts of least squares and mean value of the optimal control function to determine the number of shifts and the number of physicians needed.
Risk Management for Call Center Staffing
Jamol J. Pender
Industrial and Financial Engineering
Princeton University
In this work, we intend to explore the control of dynamic rate queueing systems. Optimal control of dynamic rate queueing systems have many applications in telecommunication systems as well as call centers. In call center staffing it is of utmost importance to staff the call center with the minimum number of agents to achieve a level of satisfaction that is acceptable for the manager. In this report we primarily look at call center from a manager and shareholder perspective. From the manager perspective, we want to staff the call center with the minimum number of agents, however, from a shareholder we want to make a consistent profit from investing in our call center. These two objectives are explored and considered in this work.
Characterizing the Impact of Risk Factors on Mortality for Breast Cancer Patients
Shengfan Zhang
Edward P. Fitts Department of Industrial and Systems Engineering
North Carolina State University
We model and compare mortality for breast cancer patients using community-based Carolina Mammography Registry data. Cumulative incidence function is used to estimate mortality probabilities from breast cancer and other causes as a function of patient age, race, cancer stage at diagnosis and breast cancer risk factors (breast density, estrogen and progesterone receptor status, and family history of breast cancer). Methods for approximating confidence intervals are also applied to enable the comparison among different risk groups. Left censoring is incorporated using a multifaceted cancer growth model to quantify the lag between actual start time of breast cancer and diagnosis time (recorded cancer start date).