This Github repository contains the code of our project where we reproduce the results of the paper "Training independent subnetworks for robust prediction" by Havasi et al. using TensorFlow.
In this project we have studied the Multi-Input-Multi-Output (MIMO) architecture in Deep Networks. The architecture was tested in several image recognition benchmarks such as Cifar10 and Cifar 100, and also on a more complex dataset that contained chest X-rays of COVID-19 patients. An additional dataset we used was the speech commands dataset, to test a different type of data besides images. We also ran some experiments on the trained networks such as testing the independence of the subnetworks:
The full results of the project can be found in the file report.pdf.
Authors: Lorenzo Mazza, Fernando Gastón and Anass Elyasini