pls give some hints on the tuning of adaptive algs

Will the adaptive algs achieve the same performance as the direct
computation of the optimal weights may achieve, especially when data
is short?
Is it necessary to estimate the variance of the input data when
initilizing the adaptive algs, like RLS?
How to monitor the stationariness and take what measure to counteract
non-stationariness?
Is there any common practices not mentioned in textbooks but necessary
to achieve rapid convergence?

On 29 Nov, 10:13, zqchen <zhiqun.c...@gmail.com> wrote:
> Will the adaptive algs achieve the same performance as the direct
> computation of the optimal weights may achieve, especially when data
> is short?

The iterative methods like LMS would likely need a long initial
transient before it stabilizes.

For short data segments, parametric models might be useful.
They probably don't qualify as "adaptive," though, and they
should not be used unsupervised.

> How to monitor the stationariness and take what measure to counteract
> non-stationariness?

The Kalman filter seems to have a lot going for it, what flexibility
and robustness is concerned. Check out the book by Durbin and
Koopman on state-space time series analysis.

Rune