Spring CSE 691 AS 5
2020 Machine Intelligence with Deep Learning 4/4/2020
Assignment 5: Convolutional Neural Network (CNN)
Description
In this assignment you will practice how to create Convolutional Neural Network (CNN) in
Tensorflow 2.0 using Keras APIs. First, you will construct CNN by follow the specific
requirements and architectures. Then, you will construct you own CNN to achieve a certain
accuracy on testing data. Both CNNs will be trained on CIFAR-10 dataset.
Instruction
In this assignment, you need to fill the block of code in the python notebook file. The
descriptions of all the functions you need to implement are as follow:
- Create CNN model (30 points): Fill codes between TO DO block to construct the
model by follow these requirements and architecture
o 32 of 5x5 filters with stride 1
o Relu activation
o 64 of 5x5 filters with stride 1
o Relu activation
o 2x2 Max Pooling with stride 2
o Fully connected with 1024 hidden neurons
o Relu activation
o Fully connected that map to 10 output classes
o Softmax
- Define optimizer and fit model (30 points):
o Adam optimizer with learning rate = 1e-5
o Show loss and accuracy
o Train for 10 epochs
o Use batch size = 128
o Split validation data from training data by using split 90:10
- Build your own model (40 points): Construct the model in TO DO block to get
the test accuracy > ~75%. You can use any available techniques.
o Initialization
o Early stop
o Optimizer
o Value of hyper parameters
o Drop out
o etc.
Note:
- In each file there are comments that walk you through the implementation, and also,
there is explanation in each block of code that you have to fill in.
- Points for each block of code is also in the comment.
- Please make sure each block of your code is runnable in Google Cloud environment.
Submission
- Your submission should contain the python notebook file
- Zip file named using the following convention:
__AS5.zip
Ex. zzhao37_ziyi_AS5.zip
- Upload zip file to blackboard before 11:59PM (EST Time) 4/12/2020