EE4305 - Mini Project - Description
Introduction
Please read the below summary for Mini Project before attempting. More detailed
instructions are provided in the ​Google Colab Notebook​.

In this project you will implement a neural network (NN) architecture from scratch. Building
the neural network will give hands-on experience converting mathematical foundations of NN
such as feed-forward and backpropagation algorithms into Python code. The project
includes the implementation of sub-functions in NN such as loss functions, activation
functions, and their derivatives. The project is divided into 3 phases:
1. Phase 1: Implement and test a baseline Neural Network
2. Phase 2: Integrate additional functionalities to the Neural Network model
3. Phase 3: Classify cancerous cells using Wisconsin Breast Cancer Dataset
Following functions/ datasets will be used to evaluate your NN models:
Classification Regression Real-world problem
AND logic Sinusoidal function Classification of cancerous
cells using ​Wisconsin
Breast Cancer Dataset​. XOR logic Gaussian function

Keypoints
● A basic code skeleton is provided for each phase. There are also plenty of hints
● The base neural network is implemented as a ​class​.
● Comments are included in each function to explain what the function does
● Codes are also provided for observation of outcomes and visualization.
● For each phase present your observations and inferences in separate text cells. You
may create additional visualizations to support your observations. For example, why and
how a parameter affects the model? Does the model overfit? Is the model too big or too
small?
Online Resources
● You may refer to online resources to help implement the model. You are also welcome
to completely rewrite the NN baseline class if you prefer to do so.
● Do not copy+paste contents. Your codes may be checked for plagiarism.
● Recommended sources: ​BP Algorithm​, ​Python Implementation 1​, ​Python
Implementation 2​, ​Python Implementation 3​.
Phases
Phase 1 (P1):​ Implement and test a baseline Neural Network
● Complete codes for the Neural Network Class: feedforward, backpropagation,
activation, loss, derivatives, train, predict, evaluate.
● Perform classification tests for AND/XOR logic.
● Perform regression tests for Sinusoidal/ Gaussian functions.
● Record results and observations
Phase 2 (P2):​ Integrate additional functionalities to the Neural Network model
● Implement codes to improve NN class (ANY 3): regularization, mini-batch training,
parameter initialization, additional layers, loss functions, activations
● Test your implementation using any dataset.
● Present your findings on how the functionality affects performance.
Phase 3 (P3):​ Classify cancerous cells using Wisconsin Breast Cancer Dataset
● Evaluate the implemented NN Class on the breast cancer dataset
● Tune the model to and present the highest accuracy score obtained
● Explain your choice of model parameters
Submission
You will have three weeks to complete the Mini Project ​(Deadline: 15 June 2020)​. Open
Google Colab notebook and make a copy from the menu (File -> save a copy in Drive). Save
the file as: “​:EE4305-Mini Project.ipynb​”.

the same file name and submit in the appropriate folder in LumiNUS (Mini Project ->
Submissions)
Instructions on using Google Colab Notebook
Colab allows you to write and execute Python in your browser, with, Zero configuration
If you have prior experience using Jupyter Notebooks, Colab is very similar. Please refer to
the ​Help menu for FAQs. ​Tools menu provides useful information on commands and
easily navigate between sections. You may also refer to ​this introductory video​.
NOTE: You will need a google account in order to access and save files under Google
Collaboratory. You may use your existing accounts in Gmail or create a new one.
Scoring
P1 - Completion of NN base class implementation
- Description of the Backpropagation algorithm
- Successful run of Classification and Regression Tests
- Observations and Inferences (Reporting)
3
1
6
4
P2 - Implementation of functions to improve model:
# 3 Methods
# 5 Methods

9
9 + 1 bonus
P3 - Successful implementation of Breast Cancer Dataset
- Observations and Inferences (Reporting)
- Accuracy for Breast Cancer test data:
# < 90% (or not obtained)
# >=90%, < 95%
# >=90%, < 95%
# >=98%
3
2

0
1
2
2 + 1 bonus
* The final score will be clipped to a maximum of 30

END OF DOCUMENT. WISH YOU ALL THE BEST

Email:51zuoyejun

@gmail.com