# Another approach is to train some algorithm using a

• Notes
• 74

Course Hero uses AI to attempt to automatically extract content from documents to surface to you and others so you can study better, e.g., in search results, to enrich docs, and more. This preview shows page 37 - 49 out of 74 pages.

Another approach is totrainsome algorithmusing a lot of labelled examplesHere are some pictures of lower-case R, and here aresome of lower-case N, you work it out!
38Supervised learningThe trendiest supervised approach at themoment is theartificial neural network (ANN)Artificial Neural Networks are a softwareconstruct loosely similar to the structure of thehuman brainThey arebiologically inspired, rather than actuallysimulating real brains
39Artificial Neural NetworksOne of the earliest versions of ANN was called theperceptron, developed in 1958A perceptron containsone neuron,which hasmany inputs but one output, and each input hasan importance, orweightTo calculate the output of a perceptron, the inputsare multiplied by their weights, and if the weightedsum is high enough, the neuron output is on
40Single-neuron perceptronSo to learn, we find the set of weights that givethe correct output for each set of inputsSum input *weightsActivation functionActivation InActivation OutActivation In
41Perceptron learningHow do you find the right weights?Start with any (random!) weights.One by one, run examples through the perceptron andchange the weights until they no longer need changingIf the output is correct don’t change the weightsIf the output is 0 and should be 1 then increase theweight by a constantdif the output is 1 and should be 0 then decrease theweight byd
42Perceptron learningPerceptrons are able to learn some simpleinput-output relationshipsAnylinearly separablerelationshipsSingle perceptrons are unable to learnfunctions like XORYou can't draw a straight line to separate the1and0output areas
43Linearly separableHere are four pointsrepresenting theORrelationship,X OR YPinkmeans the outputshould be true for thatinputA perceptron can learnthis "function"YXOnOffOffOn
45Linearly separableIt's quite a bit harderforXOR!There's nowhere toput a straight line thatwill separate theinputs that should givetruefrom those thatshould givefalseYXOnOffOffOn
Input layer46Networks of neuronsThis limitation can be resolved by usingmultiple layersof perceptrons, feeding intoeach otherABDCEwACwCEwADwBCwDEOutput layerHidden layer
47Artificial Neural NetworksThesemulti-layer perceptronsare morecommonly known as Artificial Neural NetworksThey learn in more-or-less the same way, butthe errors have topropagate backwardsthrough the network layers
48Neural NetworksOOOOO102030354045505560657080OOOOOOOOOOOOOOOOOOOOOOOfar less thanless thanabout the same asmore thanmuch more thanThe husband needs........than the wifeOOOOO90The husband is likely to beawarded......percentage of the assetshas few assetsis less than averageis about averageis wealthyis very wealthyThe marriage...

Course Hero member to access this document

Course Hero member to access this document

End of preview. Want to read all 74 pages?

Course Hero member to access this document

Term
Three
Professor
N/A
Tags
Unsupervised learning, Artificial neural network, neural network, artificial neural networks