Programme

Schedule

A provisional technical programme schedule is as follows:

  • Wednesday: 9:00 - 18:00
  • Thursday: 9:00 - 17:00
  • Friday: 9:00 - 16:00

The schedule of the meeting will provide plenty of opportunities for interaction which we hope everyone will find enriching.

Invited Plenary Talks

There will be three invited plenary talks.

Speaker: Prof. Margaret Brandeau (Stanford University, Department of Management Science and Engineering & Department of Medicine, US)

Title: How Much Detail is Enough? Examining Stochastic Elements in Models to Support Disease Control Policy

Abstract: Many potential public health policies for disease control are evaluated using epidemic models, instantiated using the best available data. Such models attempt to capture, in a stylized way, the complex stochastic interactions of individuals in a population that lead to the spread of communicable diseases. Because of data uncertainty, typical policy studies perform extensive sensitivity analysis on input parameter values. However, structural assumptions in such models, such as the choice of model type and the determination of which stochastic elements to include, might affect model predictions as much as or more than the choice of input parameters. This talk explores the potential implications of structural assumptions on epidemic model predictions and policy conclusions. We present a case study of the effects of a hypothetical HIV vaccine in multiple population subgroups over eight related transmission models, which we sequentially modify to vary over two dimensions: parameter complexity (e.g., the inclusion of age and hepatitis C virus comorbidity) and contact/simulation complexity (e.g., aggregated compartmental vs. individual/disaggregated compartmental vs. network models). We describe the findings of the case study and suggest some guidelines for future model selection. Our qualitative findings are illustrative of broader phenomena and can provide insight for modelers as they consider the appropriate balance of simplicity versus complexity in model structure.

Speaker: Prof. Kevin Glazebrook (Lancaster University Management School, Department of Management Science, UK)

Title: On Radical Extensions to Multi-armed Bandits and to Notions of Indexation

Abstract: It is nearly 50 years since Gittins (and Jones) elucidated solutions to important classes of multi-armed bandit problems (MABs) in the form of index policies. Such policies assign a calibrating index function to each option available at each decision stage and choose the option with maximal current index. There is now a huge literature related to this work and interest in MABs grows apace. The talk will discuss recent work seeking to develop appropriate notions of indexation for radical extensions to MABs. These include

  1. General models for the dynamic allocation of a single resource to a set of stochastic projects which are in competition for it. Here indices emerge as measures of the cost effectiveness of increasing the resource available to a project from a given level when in a given state;

  2. Models for optimal search in which an object is hidden in one of several locations according to a known probability distribution and the goal is to discover the object in minimum expected time by successive searches of individual locations. The work extends a classical result of Blackwell by allowing two search modes- slow and fast- to look for the object;

  3. A model for the effective sourcing of intelligence data when analytical capability is in short supply takes the form of a MAB with finite horizon in which only a small (pre-assigned) number of the bandit rewards observed may be claimed. The goal is to maximise the aggregate expected reward claimed.

In all cases an appropriate indexation emerges from a Lagrangian relaxation of the original problem.

Speaker: Prof. Kalyan Talluri (Imperial College Business School, UK)

Title: Traffic Issues for Rational Drivers

Abstract: Traffic problems and their resolution were an early preoccupation for many Operations Researchers. However, the topic has fallen on the wayside of top OR journals over the last couple of decades. The research in the area now is driven primarily by physicists and civil and traffic engineers where the modelling either has a physics flavour (to take an extreme example, the kinetic gas traffic model) or relies on discrete-event simulations to test out policies.

The advent of driverless cars and vehicle-to-vehicle communications however ought to revive interest in this problem as it has great relevance to practice and requires considerable modelling skill. In this talk we present our recent research on a simple traffic situation---a two-lane highway has one of its lanes blocked, say due to an accident. The traffic on the blocked lane has to merge to the free lane. For each car, this is akin to the classic parking problem but with a velocity decision variable, in addition to the merge decision. This can be formulated as a dynamic program and the optimal policy shown to be of a bi-threshold type. Now, however incentive compatibility comes into play. Drivers are rational and minimize their travel time. Even simple situations with just two cars on the blocked lane can result in a traffic jam (a subgame-perfect equilibrium) because of the dynamics and instantaneous best-response functions. We devise simple policies for central planner based on our insights and compare them with the optimal solutions.

(Joint work with Mihalis Markakis and Dmitrii Tikhonenko (UPF).)

Invited Tutorial

Speaker: Prof. Warren B. Powell (Princeton University, Department of Operations Research and Financial Engineering, US)

Title: Tutorial: A Unified Framework for Optimization under Uncertainty

Abstract: Stochastic optimization is a fragmented field comprised of multiple communities from within operations research (stochastic programming, Markov decision processes, simulation optimization, decision analysis), computer science (reinforcement learning, multiarmed bandit problems), engineering and economics (stochastic optimal control, optimal stopping), statistics (ranking and selection), probability (multiarmed bandit problems), and applied mathematics (stochastic search). In this talk, I will begin by presenting a much-needed canonical framework for stochastic optimization that matches the widely used setting for math programming. I will then identify the major dimensions of this rich class of problems, spanning static to fully sequential problems, offline and online learning, derivative-free and derivative-based algorithms, with special attention given to problems with expensive function evaluations. We divide solution strategies for sequential problems ("dynamic programs") between policy search (searching within a class of functions) and policies based on approximating the impact of a decision now on the future. We further divide each of these two fundamental solution approaches into two subclasses, producing four classes of policies for approaching sequential stochastic optimization problems that covers all the solution strategies that have been used in any of the fields (including whatever is currently being used in practice). We demonstrate that each of these four classes may work best, as well as opening the door to a range of hybrid policies. The goal is to create a single, elegant framework for modeling optimization problems under uncertainty, and a general tool box for designing and testing effective policies in both offline (simulated) and online (real world) settings. Every problem class, as well as the solution strategies, will be illustrated using actual applications.

Contributed Talks

Contributed talks will be split into two parallel sessions.

Social Dinner

A social dinner will be held on Thursday approx. between 18:00 - 21:00, and will take place in Ashton Hall - a 14th-century mansion recorded in the National Heritage List for England and now the Club House of Lancaster Golf Club - located 3 miles from the campus. There will be a coach service from the campus.

Ashton Hall