home · mobile · calendar · colloquia · 1995-1996 · 

Colloquium - Weigend

Modeling, Learning, and Meaning
Department of Computer Science

This talk focuses on building nonlinear models of noisy systems for predicting new instances and describing and understanding the properties underlying the system. Examples range from compilers and software engineering to music and finance.

After explaining a framework that combines the tools and insights from stochastic linear models with deterministic nonlinear models, I discuss model selection and overfitting, central for flexible models such as neural networks. I present approaches to some of the current problems in time series analysis: how to predict the volatility (local error bar) of the next forecast, how to find regimes and segment a time series based on the local noise levels, how to build multi-scale architectures that exploit different time scales of predictability, and how to extract information about the inputs and clean the time series through the emerging model.

Refreshments will be served prior to the talk.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)