Solutions to the Deep Learning course laboratories at EURECOM.
This repository contains a series of experiments built using Python and TensorFlow with the purpose of exploring modern Deep Learning techniques.
In the first notebook I developed my own implementation of a neural network using bare Python and NumPy with goal of acquiring a deep understanding of how feedforward prediction and training works.
In Lab02 notebook, instead, I worked with TensorFlow for building a convolutional neural network capable of classifying the MNIST dataset. In particular, I implemented a LeNet-5 architecture and I obtained an accuracy of 99%.
Finally, in Lab03 I have developed a Vanilla Recurrent Neural Network and a GRU (Gated Recurrent Unit) for performing Sentiment Analysis on user comments coming from several popular Internet services. At the end, I obtained a 87.5% accuracy level.