CSE 190 Lecture Notes - Lecture 2: Overfitting, Perceptron, Feature Extraction
Document Summary
Minimize j = e+ c where e is the error and c is a measure of model complexity . Have a hold out set (some fraction of the training set) Use the remaining portion to change the weights . Watch the error on the holdout set and stop when it starts to rise . Normalize brightness of the image (every image has the same brightness) Find particular feature points in the image. The perception - a single layer of processing (i. e. one layer of weights) The input to a model neuron (where w and x are the weight vector and input vector) Can be written as: either xtw or wtx. Set of weights & threshold for which 2-input perception computes or? w1 = w2 = 1, = 1. Set of weights & threshold for which 2-input perception computes and? w1 = w2 = 1, = 2. Anything it can compute, it can learn to compute! (2-layer)