Deeplearningallowscomputationalmodelsthatarecomposedofmultipleprocessinglayerstolearnrepresentationsofdatawithmultiplelevelsofabstraction.Thesemethodshavedramaticallyimprovedthestate-of-the-artinspeechrecognition,visualobjectrecognition,objectdetectionandmanyotherdomainssuchasdrugdiscoveryandgenomics.Deeplearningdiscoversintricatestructureinlargedatasetsbyusingthebackpropagationalgorithmtoindicatehowamachineshouldchangeitsinternalparametersthatareusedtocomputetherepresentationineachlayerfromtherepresentationinthepreviouslayer.Deepconvolutionalnetshavebroughtaboutbreakthroughsinprocessingimages,video,speechandaudio,whereasrecurrentnetshaveshonelightonsequentialdatasuchastextandspeech.
1