Amalgamated Modeling - Revision history

Jrising at 16:22, 13 April 2012

2012-04-13T16:22:51Z

← Older revision	Revision as of 09:22, 13 April 2012
Line 1:	Line 1:
		See the progress report presentation: [[Media:Progress_Presentation_v2.pdf‎]]

Jrising at 16:22, 13 April 2012

2012-04-13T16:22:12Z

← Older revision		Revision as of 09:22, 13 April 2012
Line 1:		Line 1:
	~~Typically, coupling two models (say, climate and oceans) requires that both be recalibrated, often with parameter values that are not internally reasonable. I'd like to find another option.~~

	My approach is to consider {X_i} to be an underlying "true" process, which is passed through a parallel collection of noisy channels, which are the models. The goal is uncover an estimate of each X_i. X_i may not be encoded before being passed through the channel.

	If the X_i and Z_i errors are independent, the solution might just be an average or a maximum likelihood. But what if the Z_i's are correlated? Suppose that the {X_i} come from a Gauss-Markov process with unknown coefficients-- can the coefficients be simultaneously estimated? Suppose that something is known about the power spectrum of X-- how can that be used? The entropy rate is a function of the spectrum, and determines the error bounds on the estimates of X_i, but what is that best estimate?

	~~[[Media:Couplingdiag.png‎]]~~

Jrising at 16:21, 13 April 2012

2012-04-13T16:21:01Z

New page

Typically, coupling two models (say, climate and oceans) requires that both be recalibrated, often with parameter values that are not internally reasonable. I'd like to find another option.

My approach is to consider {X_i} to be an underlying "true" process, which is passed through a parallel collection of noisy channels, which are the models. The goal is uncover an estimate of each X_i. X_i may not be encoded before being passed through the channel.

If the X_i and Z_i errors are independent, the solution might just be an average or a maximum likelihood. But what if the Z_i's are correlated? Suppose that the {X_i} come from a Gauss-Markov process with unknown coefficients-- can the coefficients be simultaneously estimated? Suppose that something is known about the power spectrum of X-- how can that be used? The entropy rate is a function of the spectrum, and determines the error bounds on the estimates of X_i, but what is that best estimate?

[[Media:Couplingdiag.png‎]]