Institute of Information Theory and Automation

You are here

Data dropouts in Bayesian model averaging

Date: 
2010-12-06 11:30
Room: 
Process of cold metal rolling is described by a set of linear regression models that are dynamically averaged. Weights of the models are based on predictive ability of each model in the given time instant. The models use different data sets. This property increases robustness of the model mixture: if data dropout affects only some of the models, the unaffected ones can proceed. The seminar will discuss potentialy applicable methods of adapting model weights if a dropout occurs. The methods are designed to take advantage of externally supplied additional signals informing about reliability of particular data sensors.
AttachmentSize
PDF icon utia-as-101206.pdf6.15 MB
2011-03-02 10:24