Nonlinear parameter estimation in classification problems
A nonlinear generalisation of the perceptron learning algorithm is presented and analysed. The new algorithm is designed for learning nonlinearly parametrised decision regions. It is shown that this algorithm can be viewed as a stepwise gradient descent of a certain cost function. Averaging theory is used to describe the behaviour of the algorithm, and in the process conditions guaranteeing convergence of the algorithm are established. These conditions are hard to test, so some simpler...[Show more]
|Collections||Open Access Theses|
|b19042346_Blackmore, Kim Louise.pdf||60.8 MB||Adobe PDF|
Items in Open Research are protected by copyright, with all rights reserved, unless otherwise indicated.