数据挖掘课件：CNN introduction.pptx

1、123456Introduction to Deep Learning78910 Classical CV feature definition is 11Deep NN Deep Learning promise: train good feature automatically,same method for different domainDeep NN12 Combine best of Computer Vision Deep Learning and Machine LearningCVfeaturesHoG, SIFT Deep NN13OUTPUTSHIDDEN NODESCA

2、TDOGINPUTS14CATDOG415CATDOG516CATDOG17Schmidhuber )18Convolutional Networks1920CNN - multi-layer NN architecture Convolutional + Non-Linear LayerSub-sampling LayerConvolutional +Non-L inear LayerFully connected layers Supervised Feature ExtractionClassi-fication2122232425NError-5AlgorithmTeamAuthors

3、10.153Deep Conv. Neural Network Univ. of TorontoKrizhevsky et al20.262Features + Fisher Vectors + Linear classifier ISIGunji et al30.270Features + FV + SVM OXFORD_VGG Simonyan et al40.271SIFT + FV + PQ + SVM XRCE/INRIAPerronin et al50.300Color desc. + SVM Univ. of Amsterdam van de Sande et al26NErro

4、r-5AlgorithmTeamAuthors10.117Deep Convolutional Neural NetworkClarifiZeiler20.129Deep Convolutional Neural NetworksNat.Univ SingaporeMin LIN30.135Deep Convolutional Neural NetworksNYUZeiler Fergus40.135Deep Convolutional Neural NetworksAndrew Howard50.137Deep Convolutional Neural NetworksOverfeatNYUPierre Sermanet et al2728293031323334Conv Nets:beyond Visual Classification35You need just a right nail!36373839404142434445NYU “Deep Learning” Professor LeCun Will Head Facebooks New Artificial Intelligence Lab, Dec 10, 2013