Abstract
We invoke a basic lemma from the theory of so-called penalty methods of nonlinear programming to come up with a simple yet highly efficient block-adaptive training procedure that selects a suitable Weak Continuity (WC) regularization parameter by matching the complexity (`prior') of the resulting solution to that of a desired response. The matching is achieved using a simple binary search technique that is guaranteed to converge in very few steps; as such, it avoids many of the potential pitfalls of other iterative methods.
Original language | English (US) |
---|---|
Pages (from-to) | 1009-1013 |
Number of pages | 5 |
Journal | Conference Record of the Asilomar Conference on Signals, Systems and Computers |
Volume | 2 |
State | Published - Jan 1 1997 |