详解稀疏编码（百度深度学习--ppt）

【实例简介】
百度深度学习研究院的研究介绍，稀疏编码的详解，简洁易领会，是学习稀疏编码的很好资料
The pipeline of machine visual perception Most efforts in Machine Learning OW-level Pre extract.fEature Feature terence sensing processing selection → prediction, recognition Most critical for accuracy Account for most of the computation for testing Most time-consuming in development cycle Often hand-craft in practice 7/22/2014 Computer vision features Normalized patch Spin image d=10,n ※米 d=0.0,t=10 age gradient eypo Orientation v ling RIFT Normalized patch Ce TO d=0.9,0=n4 Overlapping blocks 2.d=0.5,6=/2 Input Image Gradient Image 1.d=0.3,=丌 Local normalization 。。。。图"N oo。oolo (d) Slide Credit: Andrew Ng Learning features from data Machine Learning Inference LOW-level Pre- Feature Feature sensing processing extract selection prediction recognition Feature Learning instead of design features lets design feature learners 7/22/2014 Learning features from data via sparse coding Inference LOW-level Pre- Feature Feature sensing processing extract selection prediction recognition Sparse coding offers an effective building block to learn useful features 7/22/2014 Outline 1. Sparse coding for image classification 2. Understanding sparse coding 3. Hierarchical sparse coding 4. Other topics: e.g. structured model, scale-up discriminative training 5. Summary 7/22/2014 Bow representation SPM Paradigm-I Bag-of-visual-words representation (Bow)based on VQ coding 7/22/2014 Figure credit: Fei-Fei Li Bow representation SPM Paradigm-IT Spatial pyramid matching pooling in different scales and locations 7/22/2014 Figure credit: Svetlana Lazebnik Image Classification usingBow+SPM Image classification Dense sift (1(2),,x(m0∈R 128 VQ Coding (1)a( Spatial Pooling 讠=1 Classifier 7/22/2014 【实例截图】
【核心代码】