Three-Toed Sloth

Next Week at the Statistics Seminar: Bayes, Bayes, Baked Beans, Sausage and Bayes

Next week at the CMU statistics seminar, we give you all the Bayes you could want and more:

Tom Griffiths, "Connecting human and machine learning via probabilistic models of cognition"

Abstract: Human performance defines the standard that machine learning systems aspire to in many areas, such as forming new concepts, making scientific discoveries, and learning language. This suggests that studying human cognition may be a good way to develop better learning algorithms, as well as providing basic insights into how the mind works. However, in order for ideas to flow easily from psychology to computer science and vice versa, we need a common language for describing human and machine learning. I will summarize recent work exploring the hypothesis that probabilistic models of cognition, which view learning as a form of statistical inference, provide such a language, including results that illustrate how novel ideas from statistics can inform cognitive science.

Date and place: Monday, 13 April 2009, 4-5 pm in Porter Hall 125C

Peter Orbanz, "Conjugate Projective Limits"

Abstract: Bayesian nonparametric models are essentially Bayesian models on infinite-dimensional spaces. Most work along these lines in statistics focusses on probability models over the simplex. In machine learning, the problem has recently received much attention as well, and attempts have been made to define models on a wider range of infinite-dimensional objects, including measures, functions, and infinite permutations and graphs.

In my talk, I will discuss the construction of nonparametric Bayesian models from finite-dimensional Bayes equations, roughly analogous to the Kolmogorov extension of systems of measures to their projective limits. I will present an extension theorem applicable to regular conditional probabilities. This can be used to study whether "conditional" properties of the finite-dimensional marginal models, such as conjugacy and sufficiency, carry over to the infinite-dimensional projective limit model, and to determine the functional form of the nonparametric Bayesian posterior if the model is conjugate.

Date and Place: Friday, 17 April 2009, 1-2 pm, Porter Hall A18A

Both talks are free and open to the public.

Enigmas of Chance; Minds, Brains, and Neurons