深度学习 CNN代写 CSE 400 AS 5 2018 Machine Intelligence with Deep Learning

Fall CSE 400 AS 5 2018 Machine Intelligence with Deep Learning 11/07/2018

Homework 5: CNN using Neon

Description

In this homework you will practice how to create Convolutional Neural Network (CNN) in Python and Neon. You need to understand the concept of CNN and how to use Neon APIs.

Instruction

In this homework, you have to write 2 python files.

  • –  assignment5_1.py: Use the neon tutorial file in class as a template, you need to

    implement CNN by follow these requirements and model architecture. o Cifar10 dataset (train & test)
    o Batch size 128
    o Uniform initialization (-0.1, 0.1)

    o Model architecture
    § 32 of 5×5 filters with stride 1
    § Relu activation
    § 64 of 5×5 filters with stride 1
    § Relu activation
    § 2×2 Max Pooling with stride 2
    § Fully connected with 1024 hidden neurons
    § Relu activation
    § Fully connected that map to 10 output classes § Softmax
    § Cross entropy

    o Adam optimizer with lr=1e-5
    o Print loss value for test data every iteration o Train for 20 iterations
    o Report train & test accuracy

  • –  assignment5_2.py: Similar to assignment5_1.py, you have to implement your own model to get test accuracy > 60%.

    o Use any techniques (initialization, optimizer, value of hyper parameters, etc.) o Use Cifar10 dataset, and report train & test accuracy at the end.
    o Save your model to “myModel.prm”

    Submission

  • –  Your submission will contain 3 files assignment5_1.py, assignment5_2.py,

    myModel.prm.

  • –  Zip file named via the following convention:

    <SU-EMAIL>_<FIRST-Name>_AS5.zip

    Ex. kpugdeet_Krittaphat_AS5.zip

  • –  Upload zip file to blackboard before 11:59PM (EST Time) 11/16/2018