SP-Deep-Learning

This repository records my learning journey in the realm of Deep Learning and its subfields in Singapore Polytechnic DAAA curriculum.

⚠️ For all existing SP Students taking ST1504 Deep Learning module or any relevant Deep Learning electives, do take note that plagiarism is a serious offense and liable for necessary disciplinary actions. There have been previous incidents of students cloning/forking on this repository and have resulted to obtaining poor grades or even failing the module. Telemetry is enabled for this repository, relevant lecturers will be notified if there are spikes of viewership, cloning or forking of this repository.

Table of Contents

Note that most implementation are done on TensorFlow 2.0/Keras, despite the popularity of PyTorch. Keras was the deep learning framework of choice due to the simplicity of its design and more intuitive for lecturers to read during grading.

Title	Description
Image Classification w/ Benchmark Datasets	Build and train a convolutional network classifer in Fashion MNIST and CIFAR-10 Image Dataset. Implemented CNN architectures includes
Sentiment Analysis w/ COVID-19 Tweets	Develop and train a neural network to perform Sentiment Analysis of COVID-19 Tweets data, with a Hybrid CNN-LSTM architecture to extract and identify the affective states of the tweet.
GAN w/ CIFAR-10	Develop and train a Generative Adversarial Network (GAN) to generate synthetic images of the CIFAR-10 image dataset. Implemented GAN architecures includes GAN, Conditional GAN, and Auxiliary Classifier GAN.
RL w/ Lunar Lander	Develop and train a neural network to successfully land the Lunar Lnader successfully on the landing pad. Success condition is also denoted as $score > 200$. Implemented architectures includes DQN, DDQN, D3QN, A2C
Case Study: Supervised Binary Domain Image-to-Image Translation w. Pix2Pix cGAN	Conduct a case study on supervised Image-to-Image Translation on a paired/two-domain dataset using a conditional generative adversarial, formally called Pix2Pix. The network architecture is reproduced using the proposed model by Philip Isola, then implemented on a cityscape paired dataset. The network is trained with an objective to perform image to image translation from a semantic annotated urban street image to a realistic urban street image.
Pneumonia Classification with Explainable AI	Applications of neural networks in the Healthcare sector are rising. However, the lack of transparency in the decision-making process of the neural network is a major concern; due to the “black-box” nature of neural networks. </br> This project aims to Explainable AI methods (XAI) such as LIME and Integrated Gradients on a Pneumonia Classification model (MobileNetV3), a prediction-sensitive use-case.
_pdf	Directory containing all relevant .pdf such as slide decks and technical paper with regard to ST1504 Deep Learning module. It is suggested to view this directory for those who wants to have a quick run-through of my work.