DD2424 Deep Learning

This is a repo for the individual course assignments of DD2424 Deep Learning in Data Science at KTH 2020. The code in this repo is mainly done in Matlab, and manipulations involved in the training process like gradient calculations and parameter updates are implemented in a trival way (low level).

Table of contents

• Introduction
• Assignment01
• Assignment02
• Assignment03

Introduction

In this project, A k-layers neural networks is designed with the implemention of He Initialisation, Cyclical Learning Rates, Batch Normalisation. The objective is to use these neural network models to classfy images from the CIFAR-10 dataset.

Assignment01

• Test Accuracy: 36.96%

Simple 1-layer neural network

Assignment02

• Test Accuracy: 51.44%

2-layer neural network + Cyclical Learning Rates

Assignment03

• Test Accuracy: 53.33%

k-layer neural network + Cyclical Learning Rates + Batch Normalisation

Contact

Chieh-Ju Wu (Jeremy) - jeremy.cjwukth@gmail.com