ROMEO HPC Center

Journées DEVS francophones Deep Learning


Le Centre de Calcul ROMEO organise son école annuelle DeepLearning dans le cadre d'une session spéciale des journées DEVS francophones.

Dates des journée DEVS : du 29 avril au 5 mai 2018

Dates de l'école Deep Learning : lundi 30 avril et mardi 1er mai 2018

Formation : Deep Learning

Lieu : Institut d'Etudes Scientifiques de Cargèse - CNRS - Corse (Ajaccio)

Lien pratique et inscription : http://devs-network.org/jdf-2018

Information pratiques : les frais d'inscription incluent l'hébergement et les frais de restauration

 

 

 

 

 

Le Centre de Calcul Régional ROMEO et NVIDIA proposent une formation dédiée au Deep Learning, les lundi 30 avril et mardi 1er mai 2018 à Institut d'Etudes Scientifiques de Cargèse. Cette formation formation fait suite aux deux premières formations d'initiations et est destinée à la fois aux universitaires et entreprises qui souhaitent découvrir le Deep Learning et en expérimenter les capacités dans un environnement NVIDIA / GDX-1.

 

Une partie des sessions aura lieu en langue anglaise.

 

Pour une version numérique de l'affiche : pdf

 

Intervenants :

 



Gunter ROETH
 
Solutions Architect at NVIDIA
 

 



Greg HEINRICH
 
Deep Learning Engineer at NVIDIA

 

 

 

 

Sponsors :



 









 





 

Planning :

 

Jour 1 :

 

9h00 : Accueil / Café

9h30 : Formation : session 1

11h30 : Déjeuner

13h00 : Formaton : session 2

18h30 : fin de la session 2

19h00 : Début de Social event 

 

Jour 2 :

 

8h30 : Accueil / Café

9h00 : Formation : session 3

11h30 : Déjeuner

13h00 : Formaton : session 4

17h30 : fin de la formation

 

 

Programme :

Présentation du Centre de Calcul ROMEO

Introduction to NVIDIA software et hardware for Deep Learning

Introduction to Deep Learning

In this interactive class we will introduce the rapidly developing technology of Deep Learning accelerated by GPUs. Recent advances in Deep Learning have led to a step change in performance in a number of machine perception tasks including visual perception, speech recognition and natural language understanding after decades of slow progress in these areas. The catalyst for this progress is the advent of big data via the internet, algorithmic advances and dense computation via GPUs. We will tour the most popular software frameworks for Deep Learning with goal of helping you decide which framework best suits your application needs as a researcher or developer. No prior knowledge of Deep Learning is required.

 

Getting Started with Deep Learning

Train LeNet with MNIST and get better and better results (we have done this one)
Deep learning is giving machines near human levels of visual recognition capabilities and disrupting many applications by replacing hand-coded software with predictive models learned directly from data. This lab introduces the machine learning workflow and provides hands-on experience with using deep neural networks (DNN) to solve a real-world image classification problem. You will walk through the process of data preparation, model definition, model training and troubleshooting, validation testing and strategies for improving model performance. You will also see the benefits of GPU acceleration in the model training process. On completion of this lab you will have the knowledge to use NVIDIA DIGITS to train a DNN on your own image classification dataset.

 

Deep Learning Network Deployment

Deep learning software frameworks leverage GPU acceleration to train deep neural networks (DNNs). But what do you do with a DNN once you have trained it? The process of applying a trained DNN to new test data is often referred to as ‘inference’ or ‘deployment’. In this lab you will test three different approaches to deploying a trained DNN for inference. 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 it’s Python API. The final approach is to use the NVIDIA High PerformanceGPU Inference Engine (TensorRTGIE) 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’ll also explore inference for a variety of different DNN architectures trained in other DLI labs.


Getting Started with Caffe for Deep Learning

Caffe is a Deep Learning framework developed by the Berkeley Vision and Learning Center (BVLC) and by a large community of open-source contributors. Caffe allows the user to define, train and deploy Deep Neural Networks (DNNs) through accessible command line, Python and MATLAB interfaces. Caffe is fast due to integrated GPU acceleration. In this class we will introduce the following aspects of Caffe, demonstrated through a practical DNN training and deployment example

 

Journée Digits Object Detection et Segmentation

Torch, Approaches to Object Detection using DIGITS, Detecting Whale Faces using Object Detection and NVIDIA DIGITS

 

Introduction to Recurring Neural Networks (RNNs)

What are RNNs? - Simple example of Binary addition - RNN training with stochastic gradient descent (SGD) - Backpropagation through time (BPTT) -  Challenges such as "vanishing" and "exploding" gradients - Backprop in RNNs - Long Short Term Memory (LSTM)

 

Tensorflow with DeepDream by Google 

 

 

 

Le planning, le programme et la liste des intervenants sont en cours de construction et non définitifs.

Environnement

Chaque participant apportera son PC portable pour suivre les formations. Pensez à configurer EDUROAM depuis votre établissement.