Deep Learning Project

DO DEEP CONVOLUTIONAL NETWORKS REALLY NEED TO BE DEEP AND CONVOLUTIONAL

Repository for the DD2424 Deep Learning in Data Science course at KTH.

Authors : Samuel LEONARDO GRACIO, Victor STIMPFLING and Martin VERSTRAETE.

How to run

1. Download Jupyter Notebook or use Google colab, available at : https://colab.research.google.com
2. Download this repository.
3. Open the different ipynb files.

Note

In order to get the different saved models, should go on the link given in "link_to_saved_models.txt", which is the following link : https://drive.google.com/drive/folders/1X51jwhCpIz4fTYnFZATN8E6uHTRkf0jj?usp=sharing

Description of the project

Despite all the evidences presented by G.Urbanet al. (1) (2017), we decided to repeat empirical experiences in order to check if their conclusions are relevant for a more complex data set. To do so we choose to firstly reproduce the main experiments performed by G.Urbanet al. and secondly to adjust our network in order to match the minimal requirement in order to be performing on CIFAR-100.

Link to the original research paper

https://arxiv.org/pdf/1603.05691.pdf