Description
The goal of this assignment to understand convolutional neural networks taught in
the class.
This is an individual assignment.
You will be provided with CNN code developed in Matlab environment.
Play with the code to understand its working and performance on the given data set.
Document your observations
Submission instructions are given at the end of this document
Submission deadline: 25
[1] Download the MNIST dataset from the link: http://yann.lecun.com/exdb/mnist/ to the
distributed folder where code is present. Report the dataset details.
[2] Distributed code has one set of ‘conv-pool’ layer followed by fully connected layer.
i. Run the code (‘Traincnn’) and report the performance and training time (without
any changes in the code). Visualize the weights and report the weight patterns.
ii. Visualize the feature maps and identify which of them respond to edges or lines.
iii. Restrict your training set to first 10,000 images. Make changes to the following,
one after other and report the trained network performance in form of a table.
(3 variations each)
a. No. of epochs (‘e’ in Traincnn)
b. No. of feature maps in the convolution layer (‘numFilters’ in Traincnn)
c. Convolution layer filter dimension (‘FilterDim’ in Traincnn).
Submission Instructions
Enclose in a single zip folder the following:
1. A code to visualize the weights and feature map using the trained weights.
2. Include the trained weights.
3. A detailed report with the all the necessary images
Submit the compressed zip or tar file named as
following address. Please note do not drop your assignment on to the group.