Apply data science tools to situations encountered in physics.
Among the many classes of situations to which Data Sciences apply, we will see some examples encountered in physics. In addition to the implementation of tools developed in other courses, the course will develop various theoretical methods useful for the analysis, representation or classification of data in physics.
The course will be based on one or more situations encountered in physics whose resolution or analysis requires the use of "Data Science" methods. As an illustration, the public website Kaggle.com provides a wide variety of datasets associated with challenges related to physics problems (e.g. the detection of gravitational waves). Data from experiments conducted at UNamur will also be processed. The necessary theoretical and methodological contents, such as neural networks and deep learning, will be taught. The basics of working in Python and using libraries such as Numpy, Matplotlib, TensorFlow and Keras will be presented. One or more seminars on other aspects of the use of data sciences in physics will be offered. The student will have to solve the proposed challenge practically.
Introduction to Python
Working with Anaconda (Python 3.x)
Installation of libraries
The Numpy library for scientific calculation
The Matplotlib library for graphics
TensorFlow & Keras
Installation of TensorFlow (CPU & GPU)
Installation of Keras
Working with TensorFlow & Keras on the cluster
Challenge & Datasets
Logistic regression
Neural networks I
Deep Neural Networks (DNN) : theory & implementation with Keras
Exercise on logic functions
DNN on MNIST
Neural Networks II (advanced implémentation)
Matricial formulation & back-propagation algorithm
Batch Gradient Descent with Momentum, Adam, RMS Prop
Regularization techniques
The concepts of entropy and cross-entropy
Tuning of hyperparameters
DNN on MNIST revisited
DNN with visualization of features
Convolutional Neural Networks
CNN on MNIST
CNN with visualization of features
Pretrained Convolutional Neural Networks
VGG16 on MNIST
VGG16 with visualization of features
Neural Style Transfer
Deep Dreaming with Inception
PCA, t-SNE, AutoEncoders & Variational AutoEncoders
PCA, t-SNE, AutoEncoders on MNIST
Variational AutoEncoders on MNIST
Variational AutoEncoders & Art
Advanced thematics
The ex-cathedra courses will be complemented by the use of online resources and seminars by guest speakers. Students will be monitored individually for the practical resolution of challenges.
A written exam will evaluate the theoretical concepts of the course (8 points). The student will demonstrate his ability to apply the concepts of the course (essentially convolutional neural networks with appropriate regularization techniques) to the classification of a dataset of his choice (8 points). The work done during the practical sessions counts for 4 points.
| Training | Study programme | Block | Credits | Mandatory |
|---|---|---|---|---|
| Master in Physics | Finalité spécialisée en physique et data | 1 | 3 | Yes |