DeepLearning Spring2018

---

Introduction to Deep Learning Labs

Collection of Deep Learning approaches, starting from simple Neural Network architectures and pursuing with contemporary and state-of-the-art practices and models.

This repository contains Jypyter notebook python codes from the course Deep Learning taught at EURECOM, by prof. Benoit Huet, in Spring 2018.

The 3 labs were:

1. Artificial Neural Networks (ANN): Handwritten Digits Recognition problem (yann.lecun.com/exdb/mnist/) with different settings of hyperparameters. NNet class implementation from scratch, allowing for flexible definition and training/evaluation of conventional neural networks (MLP).
2. Convolutional Neural Network (CNN) for Handwritten Digits Recognition: CNNet function implementation based on Tensorflow, implementing a generalized model to use such library as well as the LeNet5 model.
3. Sentiment Analysis with Recurrent Neural Networks (RNN): using a Vanilla RNN and a GRU (Gated recurrent unit) on TensorFlow

All notebooks follow a general structure:

1. Data import and descriptive analytics (often using the MNIST dataset)
2. Implementation of a somewhat simple model and initial testing
3. Implementation of the advanced model and exhaustive testing
4. Parameter tuning and further improvements

Details

• Language: Python over Jupyter Notebooks.
• Execution: set over a remote Spark cluster in EURECOM, managed by Zoe
• Libraries: tensorflow, numpy, matplotlib

Authors

• Albane Bonnaud
• Kim-Anh-Nhi Nguyen @kanguyn

Motivation

Deep Learning is a new approach in Machine Learning which allows to build models that have shown superior performance fora wide range of applications, in particular Computer Vision and Natural Language Processing. Thanks to the joint availability of large data corpus and affordable processing power, Deep Learning has revived the old field of Artificial Neural Networks and provoked the "Renaissance" of AI (Artificial Intelligence). The present notebooks will address this topic, implementing from scratch and using well known libraries, Deep Neural Networks.

Sources and acknowledgments

The initial assignments come from the course in DeepLearning taught in EURECOM by Benoit Huet

The notebooks have been crafted mainly by their authors, following well known open source documentation for the different libraries used and the material from the course.

The notebooks are based on publicly available data.

License

MIT Free software