Special PDE Seminar
*PLEASE NOTE CHANGE OF DAY, TIME AND PLACE FOR THIS WEEK ONLY |
Brown University Center for Statistical Sciences Seminar
The Cleveland Clinic Foundation | |
First Floor Conference Room, Dean's Suite * Refreshments Following Seminar at 167 Angell Street, 2nd Floor Conference Room. |
Abstract: In this talk I will discuss three issues in the design and analysis of diagnostic accuracy studies. First, the conventional ROC curve and associated indices assume two truth states. However, clinicians must often interpret tests to distinguish between multiple truth states. I propose a new format for collecting readers' interpretation data (called the ``differential diagnosis'') which can be used to estimate pair wise measures of accuracy between various truth states, as well as a summary measure of test accuracy. Second, with the development of new noninvasive and less expensive tests, we are often interested in showing that a test has equivalent diagnostic accuracy to a standard test. I propose two criteria for evaluating the equivalence of tests' diagnostic accuracies. These criteria look at the distribution of readers' accuracies, not just the average accuracy, and use the inherent intra-and inter-reader variability on the standard test as a benchmark for comparing the two tests. Finally, prospective studies with low prevalence rates can require and enormous sample of patients. I investigate two strategies for reducing the sample size of such studies; stratification and two-stage sampling. These strategies can be cost-effective in various settings and have complementary roles.
Special Applied Mathematics Colloquium - TURKEY TALK
and The Romance of 39615 | |
Special Lefschetz Center for Dynamical Systems Seminar
Special Brown Applied Mathematics Pattern Theory and Vision Seminar
Abstract: Neural responses have long been characterized by the number of action potentials (spikes) over some long interval (seconds), averaged over many repeated presentations of a fixed stimulus or behavioral condition. More recently, some neurobiologists have begun to describe neural responses at the level of individual spikes at millisecond resolution, and to quantify the variability in timing and number of spikes from one presentation to the next. There has been considerable debate about whether the original ``rate code'' description was sufficient, or whether individual neurons use ``temporal codes'', but there has been no consensus about what would constitute a rate code or temporal code. Most proposed definitions have confounded two issues: the temporal precision with which spikes are fired, and the temporal interactions among spikes. I will describe a proposed information-theoretic measure that I claim differentiates these concepts. This measure compares the information content in a spike train to the estimate that would be obtained on an assumption of temporal independence. In this talk I will focus on the properties of this measure: its relation to other quantities, its performance on real and artificial spike codes, and some remaining deficiencies and open questions.
<--- 2000 Index