Intelligence Artificielle et Multimédia
Objectives
At the end of this module, the student should have understood and be able to explain (main concepts):
- The use of deep learning algorithms for the classification of complex high-dimensional data with prediction error estimation
- The main classification algorithms for media data
- Applications of deep learning methods to real-world data sets
The student should be able to:
- Fit deep neural networks for classification or regression of media data: images, videos, 3D.
- Implement deep learning algorithms on real data using Python libraries.
Description
This course is dedicated to learning methods and in particular deep learning methods, for the processing of high dimensional data such as images for example.
- Neural networks and introduction to deep learning: definition of neural networks, activation functions, multilayer perceptron, gradient backpropagation algorithm, optimization algorithms, regularization.
- Convolutional neural networks: convolutional layer, pooling, dropout, architecture of convolutional networks, learning transfer, applications to image classification, object detection, image segmentation, posture estimation, etc.
- Recurrent neural networks: sequence modeling, recurrent neuron, backpropagation through time, LSTM and GRU, applications to natural language processing and audio and video signal processing.
- Neural networks and 3D: 3D convolutional networks for volumetric data processing (e.g. MRI), PointNet and PointNet++ networks for 3D point cloud processing (e.g. LIDAR).
Targeted skills
deep learning, deep neural networks
Bibliography
- Goofellow I., Bengio Y., Courville A. “ Deep Learning”, MIT Press
- Hastie, T. Tibshirani, R., Friedman, J. “The elements of statistical learning”, Springer, 2001
- Chollet, F. “Deep Learning with Python”, Manning Publications, 2018
Pre-requisites
Statistics, Statistical learning