About

SUTD 50.039 Theory and Practice of Deep Learning Coursework. For more information, refer to https://istd.sutd.edu.sg/undergraduate/courses/50-039-theory-and-practice-of-deep-learning.

Material

Homework

1. HW1 [Code] - sklearn Support Vector Classification (SVC)

Using sklearn Support Vector Classification (SVC) to implement a linear SVM. The dataset used is imageclef2011.

2. HW2 [Code] - Part 1 (Broadcasting), Part 2 (k-means)

Part 1 (Broadcasting) - Comparing time measurements using double for-loops, numpy broadcasting and pytorch cpu
Part 2 (k-means) - Creating a pytorch k-means algorithm

3. HW3 [Code] - Part 1 (SGD for logistic regression in Pytorch), Part 2 (Fully connected neural net in Pytorch)

Part 1 - Implement SGD for logistic regression using a self generated dataset.
Part 2 - Implement a fully connected neural network (FCNN) using the FashionMNIST dataset.

4. HW4 [Code] - Part 1 (Data augmentation), Part 2 (Finetuning neural network)

Part 1 (Data Augmentation) - center crop testing, 5-crop testing, varying input size
Part 2 (Finetuning neural network) - Finetuning a neural network using the 102-multiclass flower dataset available for download at https://www.kaggle.com/demonplus/flower-dataset-102
(a) training all layers with no pretrained weights
(b) training all layers with pretrained weights
(c) training only last two layers with pretrained weights

5. HW5 [Code] - LSTM and GRU to predict the language of names

Using RNNs (LSTM/GRU) to predict the language of names