Stochastic Information Gradient

From ITL Wiki
Jump to: navigation, search

Learning Curves for MEE-SIG

A two tap adaptive filter was used to identify another two tap FIR filter. The input to both the filters was white Gaussian noise with unit power. The step size of the MEE-SIG weight update rule was set to 1. The figures below show the learning curves of the adaptive filter, for different values of kernel size, and also for different values of the batch size L, used for computing the stochastic gradient:

Sig sigma.png

Sig l.png

Personal tools
EEL 6935