Brown University Center for Statistical Sciences Seminar
The Wharton School of Business, University of Pennsylvania | |
1st Floor Conference Room |
Abstract: Fisher (1935) described randomization as the reasoned basis for inference in experiments, for it created without assumptions the required distributions. Fishers framework encourages randomized experiments when ethical and feasible, and it sharpens discussion of the issues that arise in observational studies when randomization is not possible. A limitation is that many randomization tests are not paired with confidence intervals. Randomization-based confidence intervals are well known for additive treatment effects, but many effects, including effects for binary responses, are not additive. Attributable effects are a fairly general device for inverting randomization tests to obtain confidence statements in experiments, and to perform sensitivity analyses in observational studies. The method will be discussed with several examples.
*Co sponsored by PSTC
Cognitive & Linguistic Sciences Colloquium Series
Abstract: Usage-based approaches to language have gained great popularity in a variety of domains, including theoretical linguistics, natural language processing, child language acquisition, and computer modeling. The evidence that language knowledge and language use are extremely sensitive to the fine details of experience is quite persuasive, and compatible with a number of statistically-based learning methods.
Nonetheless, there are a number of important questions and outstanding problems to be considered: What are the limits to statistically-based learning? Do language users literally record their experience in some numerical form? How do we explain abstraction and generalization beyond literal experience? How closely tied to experience should we expect language processing to be? In short, what is the nature of generalization in such models?
These questions will be the focus of my talk. I shall argue in favor of usage-based language knowledge, but I shall also argue for the importance of constraints. The crucial question is what is the nature of these constraints: Are they domain- specific to language, or do they reflect constraints from other sources that intersect to yield a language-specific outcome?
Refreshments will be served before the talk in Room 124-125 Please contact Christie Crozier if you would like to meet with Dr. Elman
Brown University
Graduate School
Dissertation Defense Information
Brown Applied Mathematics Pattern Theory and Vision Seminar
-- Mark Johnson and Eugene Charniak -- |
Abstract: Partial utterances, restarts and repairs are just some of the disfluencies that are present in spontaneous speech. This talk focuses specifically on restarts and repairs, such as: I want a flight to Denver, um, to Boston on Friday.
One of the challenges in modeling speech repairs is that while their distribution certainly is related to syntactic structure, they usually don't align neatly with a syntactic structure. In fact, the most natural way to model repairs is in terms of a kind of "rough copy" structure that is orthogonal to standard syntactic structure, and, in fact seems to involve non-context-free types of dependencies. The noisy channel model seems to be a natural way of simultaneously modeling the syntactic structure and the repair structure of an utterance. This talk describes a noisy channel model of speech repairs that uses a stochastic parsing model as a model of syntactic structure and a channel model based on Tree Adjoining Grammars to model the repairs.
Brown Analysis Seminar
Special Lecture Series
Special Graduate Student Seminar
Abstract: Starting with very simple things, various methods will be described for proving existence of solutions of nonlinear problems. These will include continuity method; degree theory; fixed point theory; some variational methods.
This is the Special Graduate Student Seminar of the Mathematics Department.
Special Lecture Series
Scientific Computing Seminar
and Institute of Applied Mathematics (IAM), University of British Columbia, Canada | |
Abstract: In this talk, we present a new and robust reconstruction method called the inverse polynomial reconstruction method as a way of removing the Gibbs oscillations that are contained in the approximations of discontinuous functions or in the Fourier approximations of nonperiodic functions.
First, we present the inverse reconstruction from the Fourier approximation of an analytic nonperiodic function on a finite interval. The inverse method reconstructs a function as a polynomial such that the residue between the Fourier representations of the reconstructed function and the original function is orthogonal to the Fourier or polynomial space. As a result, the method is reduced to linear systems and requires the inversion of the transfomation matrix. We show that the reconstruction with the inverse method is uniquely determined independently of the basis sets used for the reconstruction and that it is exact if the original function is a polynomial. Furthermore, we show that the inverse method recovers spectral convergence as the truncation error and the regularization error have the same magnitude in the Fourier space.
Then we generalize the method to the case of polynomial approximations of a discontinuous function. We show that the reconstruction can be done with any polynomial basis sets and there is no conditions on choosing the basis sets for spectral convergence. We present some numerical examples to show the highly accurate performance of the method.
Finally we explain some numerical issues of the inverse method such as ill-posedness and the consistency of the transformation matrix.
*This is a joint work with Bernie Shizgal at University of British Columbia.
<--- 2004 Index