e- A face recognition dataset, with 10 persons, and 10 training images per person each with...

Transcribed Image Text:

e- A face recognition dataset, with 10 persons, and 10 training images per person each with size (52-by-52) is given. Both a deep auto-encoder with a (2704-1000-500-90) structure is compared with a CNN with 3 convolution layers, 10 filters per layer (each filter is 5-by-5), where each convolution layer uses the (valid option). Each Convolution Layer is followed by Relu anda (2-by-2, stride 2) pooling. Both the AE and CNN have a supervised fully-connected output layer then a Softmax layer. (v) Sketch both the auto-encoder (AE) and the CNN showing details of each layer (vi) How many total weights are used in the AE (vii) How many weights in each convolution layer of the CNN? (viii) How many total neurons in the AE and the CNN? (ix) Explain the steps of training both the AE and the CNN e- A face recognition dataset, with 10 persons, and 10 training images per person each with size (52-by-52) is given. Both a deep auto-encoder with a (2704-1000-500-90) structure is compared with a CNN with 3 convolution layers, 10 filters per layer (each filter is 5-by-5), where each convolution layer uses the (valid option). Each Convolution Layer is followed by Relu anda (2-by-2, stride 2) pooling. Both the AE and CNN have a supervised fully-connected output layer then a Softmax layer. (v) Sketch both the auto-encoder (AE) and the CNN showing details of each layer (vi) How many total weights are used in the AE (vii) How many weights in each convolution layer of the CNN? (viii) How many total neurons in the AE and the CNN? (ix) Explain the steps of training both the AE and the CNN

Answer rating: 100% (QA)

v Sketch of AutoEncoder AE Input 52x52 v AE Layer 1 2704 neurons v AE Layer 2 1000 neurons v AE Layer 3 500 neurons v AE Layer 4 90 neurons v Fully Co... View the full answer