Reference no: EM133385254
Question 1. The aim of this question is to perform an ablation study for the depth of the network in the program shared in notebook Conv_neural_network.ipynb.
• Report the accuracy values for Experiment-1 and Experiment-2, where the experiments add new Conv2D layers for filters and strides as described in the Table below. Note that kernel_size = 3 and padding = 'same' for new Conv2D layers
• Do you observe any difference in performance by modifying Conv2D layers? If yes then provide justification.
Question 2. The aim of this question is to perform an ablation study for the effects of batch normalization.
• Report the accuracy values for Experiment-1 and Experiment-2, where above experiments modify the BatchNormalization layers as shown in Table below.
• Do you observe any difference in performance between the above experimental setups? If yes then provide justification.
Question 3. The aim of this question is to perform an ablation study for the effects of Dropout layers.
• Report the accuracy values for Experiment-1, Experiment-2, and Experiment-3, where the experiments modify the Dropout layers as shown in Table below
• Do you observe any difference in performance between the above experimental setups? If yes then provide justification.
Question 4. The aim of this question is to perform an ablation study for the Inception Block (Experiment 1) and ResNet Block (Experiment 2) in the program shared in notebook Conv_neural_network.ipynb.
• Report the accuracy values for Experiment-1 (Inception Block) and Experiment-2 (ResNet Block).
• Do you observe any difference in performance by using Inception Block and ResNet Block? If yes then provide justification
Attachment:- Neural_network.rar