Learning rule wi k y y xi where desired output

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: hts: w1, w2, w3, … Each input pattern is classified into one of two classes depending on whether y > 0 or y < 0 . Learning rule: Δwi = k ( y − y ) xi where desired output: y learning rate: k > 0 Example of supervised learning Barn Owl Behavior: Visual input serves as a teacher for learning to orient towards an auditory target. Neurophysiology: Some neurons in the optic tectum have both a visual receptive field (V) and an auditory receptive field (A). Prisms shift the location of visual space, making it misaligned with the auditory space. After training, the auditory receptive field is shifted to realign with the visual receptive field. Supervised learning: Relation to optimal linear mapping Optimal linear mapping y = Wx finds the weight matrix W that minimizes M E=∑y ( m) m =1 − Wx ( m) 2 where ⎡ x (1) ,, x ( M ) ⎤ = X are the input vectors, and ⎣ ⎦ ⎡ y (1) ,, y ( M ) ⎤ = Y are the desired output vectors. The result is W = YX † ⎦ ⎣ where X † is the pseudoinverse of X. Perceptron learning rule for y = Wx is ΔW = k ( y − y )x T ("online learning")....
View Full Document

This document was uploaded on 02/28/2014 for the course BME 580.402 at Johns Hopkins.

Ask a homework question - tutors are online