<html><body><div style="font-family: garamond,new york,times,serif; font-size: 12pt; color: #000000"><div data-marker="__QUOTED_TEXT__"><div style="font-family: garamond,new york,times,serif; font-size: 12pt; color: #000000"><div>Yingfei Wang will present her FPO, "Advances in decision-making under uncertainty: inference, finite-time analysis, and health applications" on Tuesday, 5/9/2017 at 1pm in CS 302.<br><br></div><div>The members of her committee are: </div><div>Adviser: Warren Powell (ORFE)</div><div>Readers: Bernard Chazelle and Mengdi Wang (ORFE)</div><div>Examiners: Szymon Rusinkiewicz, Han Liu (ORFE) and <span style="color: #000000; font-family: garamond, 'new york', times, serif; font-size: 16px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; orphans: 2; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial; display: inline !important; float: none;">Warren Powell (ORFE)</span><br><br>A copy of her thesis is available in Room 310. Everyone is invited to attend her talk. The talk abstract follow below:</div><br><div>This thesis considers the problem of sequentially making decisions under uncertainty,<br>exploring the ways where efficient information collection influences and improves<br>decision-making strategies. Most previous optimal learning approaches are restricted<br>to fully sequential settings with Gaussian noise models where exact analytic solutions<br>can be easily obtained. In this thesis, we bridge the gap between statistics, machine<br>learning and optimal learning by providing a comprehensive set of techniques<br>that span from designing appropriate stochastic models to describe the uncertain environment,<br>to proposing novel statistical models and inferences, to finite-time and<br>asymptotic guarantees, with an emphasis on how ecient<br>information collection can expand access, decrease costs and improve quality in health care.<br>Specifically, we provide the first finite-time bound for the knowledge gradient<br>policy. Since there are many situations where the outcomes are dichotomous, we<br>consider the problem of sequentially making decisions that are rewarded by “successes”<br>and “failures”. The binary outcome can be predicted through an unknown<br>relationship that depends on partially controllable attributes of each instance. With<br>the adaptation of an online Bayesian linear classifier, we design a knowledge gradient<br>(KG) policy to guide the experiment. Motivated by personalized medicine where a<br>treatment regime is a function that maps individual patient information to a recommended<br>treatment, hence explicitly incorporating the heterogeneity in need for<br>treatment across individuals, we further extend our knowledge gradient policy to a<br>Bayesian contextual bandits setting. Since the sparsity and the relatively small number<br>of patients make leaning more dicult, we design an ensemble optimal learning<br>method, in which multiple models are strategically generated and combined to minimize<br>the incorrect selection of a particularly poorly performing statistical model.<br>Driven by numerous needs among materials science society, we developed a KG policy<br>for sequential experiments when experiments can be conducted in parallel and/or<br>there are multiple tunable parameters which are decided at di↵erent stages in the<br>process. Finally, we present a new Modular, Optimal Learning Testing Environment<br>(MOLTE) as a public-domain test environment to facilitate the process of more<br>comprehensive comparisons, on a broader set of test problems and a broader set of<br>policies.</div></div><br></div></div></body></html>