LeNet-5 Convolutional Neural Network Architecture
July 25, 2019
Architecture of LeNet-5, a Convolutional Neural Network, in this paper used for handwritten digit recognition. Each plane is a feature map, i.e. a set of units whose weights are constrained to be identical. LeNet-5 comprises 7 layers, not counting the input, all of which contain trainable parameters (weights). The input is a 32x32 pixel image. This is significantly larger than the largest character in the (MNIST) database (at most 20x20 pixels centered in a 28x28 field). The reason is that it is desirable that potential distinctive features such as stroke end-points or corner can appear in the center of the receptive field of the highest-level feature detectors. In LeNet-5 the set of centers of the receiptive fields of the last convolutional layer form a 20x20 area in the center of the 32x32 input. The values of the input pixels are normalized so that the background level (white) corresponds to a value of -0.1 and the foreground (black) corresponds to 1.175. This makes the mean input roughly 0, and the variance roughly 1 which accelerates learning.
Source: LeCun, Yann, et al. "Gradient-based learning applied to document recognition." Proceedings of the IEEE 86.11 (1998) 2278-2324.
Source URL: http://yann.lecun.com/exdb/publis/index.html#lecun-98
ID: lenet-5-convolutional-neural-network-architecture
Categories:
Tags: Boerhaave | COGWEB | Chinese | LSTM | Leiden | PGM | RNN | Ruysch | actors | aesthesis | agency | algorithm | analysis | anatomical | anatomy | androgynous | architecture | archive | artificialia | axis | black-box | body | botanical | brain | categories | categorization | channel | character recognition | chinese | classification | clustering | cnn | codes | cognition | collecting | collection | collections | colonialism | commodification | concept | conceptual-clustering | convolutional neural network | cost | counting | cut | cuts | cutting | datasets | demonstration | diagram | dimensionality | disgust | dissection | distance | domestication | elegance | epistemology | error | euclidean | evaluation | eye | figures | finger | forecasting | forensics | frame | freakish | geometry | gesture | gestures | gradient descent | graph | graphs | grouping | hacking | hand | hand writing | hands | hands-on | handwriting | hardware | history | human | human body | imagination | imperfect | inscription | instruments | joint | kmeans | knowledge | labeling | landmark | learning | location | machine learning | machines | materiality | meaning | measurement | memory | mnist | model | models | monsters | muscles | mystical | mythological | naturalia | nerves | nervous system | network | networks | neural networks | neural-anatomy | neuron | nonlinearity | observation | offline | online | ontologies | ontology | ontology-building | optimization | orientation | orthogonality | parallel | pca | perception | perceptron | perfection | performance | planes | poetic | position | prediction | preparation | preparations | projection | proportion | proportions | psychology | python | races | representation | representations | rhetoric | rnn | segments | selection | sensory experience | sensory perception | similarity | skeleton | skin | skull | skulls | space | sparseness | spectacle | spectators | speech | standard | statistic-ontology | statistical | statistical-ontology | svm | symbols | tacit | taxonomy | theatre | time-series | timeseries | tools | topological | training | treatise | trial | truth | type | typography | unsupervised | vision | visualization | wellcome | word2vec | writing | zodiac |