UCA Deep Learning School

Présentation

Vous pouvez choisir d'assister à tout ou partie de cette école. Merci de vous enregistrer en fonction, en suivant le lien en bas de cette page. Le nombre de places est limité !

Chaque journée est organisée en une matinée de cours/tutoriels et présentations de résultats scientifiques récents, puis en une après-midi de mise en œuvre et d’implémentation sous la direction d’instructeurs NVidia.

Une pause-café sera organisée par demi-journée.

Pour les Labs NVidia, il faut suivre les recommandations logicielles et matérielles indiquées sur le lien au bas de la page.

LIEU : Amphi du Bâtiment Forum, Campus SophiaTech
(Les horaires et les programmes sont susceptibles de changements mineurs).

Lundi 12 Juin

Lundi 12 Juin Matin (9h-12h30)

  • Panorama du Deep Learning aujourd'hui : Stéphane Canu (Deep_In_France, INSA-Rouen)
  • Bases du Deep Learning (Perceptron, MLP, backpropagation, etc) : Frédéric Precioso (Deep_In_France, UCA)
  • Introduction aux Convolutional Neural Network (CNN) : Frédéric Precioso (Deep_In_France, UCA)

Lundi 12 Juin Après-Midi (13h30-16h)

  • NVidia Lab: Getting started - Image classification on DIGITS (classification of written numbers from 0 to 9)
    • Learn how to leverage deep neural networks (DNN) within the deep learning workflow to solve a real-world image classification problem using NVIDIA DIGITS. You will walk through the process of data preparation, model definition, model training and troubleshooting. You will use validation data to test and try different strategies for improving model performance using GPUs. On completion of this lab, you will be able to use NVIDIA DIGITS to train a DNN on your own image classification application.
  • NVidia Lab: Object Detection
    • This lab explores three approaches to identify a specific feature within an image. Each approach is measured in relation to three metrics: model training time, model accuracy and speed of detection during deployment. On completion of this lab, you will understand the merits of each approach and learn how to detect objects using neural networks trained on NVIDIA DIGITS on real-world datasets.

Mardi 13 Juin

Mardi 13 Juin Matin (9h-12h30)

  • Cours autoencodeurs et Restricted Boltzman Machines : Soufiane Belharbi (Deep_In_France, INSA-Rouen)
  • Generative Adversarial Networks : Mélanie Ducoffe (Deep_In_France, UCA)
  • Transfer learning with CNN : Soufiane Belharbi (Deep_In_France, INSA-Rouen)

Mardi 13 Juin Après-midi (14h-17h30)

  • NVidia Lab: Medical Image Segmentation
    • This lab explores various approaches to the problem of semantic image segmentation, which is a generalization of image classification where class predictions are made at the pixel level. In this lab, we will use the Sunnybrook Cardiac Data to train a neural network to learn to locate the left ventricle on MRI images. In this lab, you will learn how to use popular image classification neural networks for semantic segmentation, how to extend Caffe with custom Python layers, become familiar with the concept of transfer learning and you will get to train two neural networks from the family of Fully Convolutional Networks (FCN).
  • NVidia Lab: Generative Adversarial Networks
    • Details to come…

Mercredi 14 Juin

Mercredi 14 Juin Matin (9h-12h30)

  • Cours sur les Reccurrent Neural Network: Rémi Cadène (Deep_In_France, LIP6)
  • Visual Question Answering: Rémi Cadène (Deep_In_France, LIP6)
  • Attention model pour image captioning: Jakob Verbeek (Deep_In_France, INRIA Rhone-Alpes)

Mercredi 14 Juin Après-midi (14h-17h30)

  • NVidia Lab : Modelling Time Series Data with Recurrent Neural Networks / Deep Learning with Electronic Health Records
    • The primary purpose here is to explore how deep learning can be leveraged in a healthcare setting to predict severity of illness in patients based on information provided in electronic health records (EHR). In this lab we will use the python library pandas to manage dataset provided in HDF5 format and deep learning framework keras to build recurrent neural networks (RNN). In particular, this lab will construct a special kind of deep recurrent neural network that is called a long-short term memory network (LSTM). The general idea here is to develop an analytic framework powered by deep learning techniques that provides medical professionals the capability to generate patient mortality predictions at any time of interest. Such a solution provides essential feedback to clinicians when trying to assess the impact of treatment decisions or raise early warning signs to flag at risk patients in a busy hospital care setting. Finally, we will compare the performance of this LSTM approach to standard mortality indices such as PIM2 and PRISM3 as well as contrast alternative solution formulations using more traditional machine learning methods like logistic regression. performance against baseline data.

Jeudi 15 Juin

Jeudi 15 Juin Matin (9h-12h30)

  • Optimisation Deep Networks: Mélanie Ducoffe/Soufiane Belharbi/Frédéric Precioso (Deep_In_France)
  • Convolutional neural fabrics: Jakob Verbeek (Deep_In_France, INRIA Rhone-Alpes)
  • Active learning for Deep Nets: Mélanie Ducoffe (Deep_In_France, UCA)

Jeudi 15 Juin Après-midi (14h-17h30)

  • NVidia Lab: Signal Processing using DIGITS
    • There are many resources available for learning how to leverage Deep Learning to process imagery. However, very few resources exist to demonstrate how to process data from other sensors, such as acoustic, seismic, radio or radar. In this tutorial, we will introduce some basic methods for utilizing a Convolutional Neural Network (CNN) to process Radio Frequency (RF) signals. More specifically, we will look at the classic problem of detecting a weak signal corrupted by noise. We will show you how to leverage the DIGITS application to read in a dataset, train a CNN, adjust hyper-parameters and then test and evaluate the performance of your model.
  • NVidia Lab: Deep Learning Neural Network Deployment
    • Once a deep neural network (DNN) has been trained using GPU acceleration, it needs to be deployed into production. The step after training is called inference, as it uses a trained DNN to make predictions of new data. This lab will show three approaches for deployment. The first approach is to directly use inference functionality within a deep learning framework, in this case DIGITS and Caffe. The second approach is to integrate inference within a custom application by using a deep learning framework API, again using Caffe, but this time through its Python API. The final approach is to use the NVIDIA TensorRT™, which will automatically create an optimized inference run-time from a trained Caffe model and network description file. You will learn about the role of batch size in inference performance, as well as various optimizations that can be made in the inference process. You will also explore inference for a variety of different DNN architectures trained in other DLI labs.

Inscriptions

Nous vous rappelons qu’en raison du plan vigipirate attentat et des normes en vigueur dans notre établissement l’inscription à cet événement est obligatoire.

Accéder à la plateforme d'inscription

++resource++unice.templates.stylesheets/onepage/hero75.css

++resource++unice.templates.stylesheets/onepage/hero_image_only.css