10 1 Widrow-Hoff Learning (LMS Algorithm)
10 2 ADALINE Network w i w i1 w i2 w iR =
10 3 Two-Input ADALINE
10 4 Mean Square Error Training Set: Input:Target: Notation: Mean Square Error:
10 5 Error Analysis The mean square error for the ADALINE Network is a quadratic function:
10 6 Stationary Point Hessian Matrix: The correlation matrix R must be at least positive semidefinite. If there are any zero eigenvalues, the performance index will either have a weak minumum or else no stationary point, otherwise there will be a unique global minimum x*. If R is positive definite:
10 7 Approximate Steepest Descent Approximate mean square error (one sample): Approximate (stochastic) gradient:
10 8 Approximate Gradient Calculation
10 9 LMS Algorithm
10 Multiple-Neuron Case Matrix Form:
10 11 Analysis of Convergence For stability, the eigenvalues of this matrix must fall inside the unit circle.
10 12 Conditions for Stability Therefore the stability condition simplifies to 12 i –1– Since,. (where i is an eigenvalue of R)
10 13 Steady State Response If the system is stable, then a steady state condition will be reached. The solution to this equation is This is also the strong minimum of the performance index.
10 14 Example BananaApple
10 15 Iteration One Banana
10 16 Iteration Two Apple
10 17 Iteration Three
10 18 Adaptive Filtering Tapped Delay LineAdaptive Filter
10 19 Example: Noise Cancellation
10 20 Noise Cancellation Adaptive Filter
10 21 Correlation Matrix
10 22 Signals 1.2 cos0.36–==mk k – sin=
10 23 Stationary Point 0 0 h Esk mk + vk Esk mk + vk1– =
10 24 Performance Index
10 25 LMS Response
10 26 Echo Cancellation