Speakers

The goal of the talk is to present what a ServiceMesh is, what kind of value does it brings to a distributed system.

To illustrate each value, examples will be provided on top of the leading project in that area: Istio.

The goal of the lab is to have people hands on kubernetes, manipulate core concepts such as pod, deployment, configmap, services (labels and selector mechanism), endpoints, job. From a pre installed application we will guide people th rough various use case, from application setup and rollout to investigation. This lab could be considered a learning by practice session. Attendees’ laptops will be used to connect to a prepared kubernetes instance.

The following core concepts will be described and manipulated: Deployment Pod Container configuration Volumes Probes ConfigMap Services Labels and Selector Endpoints Job.

Attendees will have to join with their own laptop and will have to connect to wifi. The lab will happen online on a preprovisioned kubernetes cluster.

The lab will be animated by David Benque and Cedric Lamoriniere.

In this workshop, you will discover many things about MongoDB : – JSON modelisation, – CRUD – Import / export – Indexes – Aggregation framework – MongoDB Driver – Replica Set – Sharded Cluster – MongoDB Atlas – MongoDB Compass – Change Stream – Security basics No existing knowledge is required.

Notes :

This is a hands-on session. People are expected to bring their laptop without any special requirements except an Internet access.

This workshop will show you how Git can be learned and understood step by step. It will train you to become familiar with the most useful Git commands and the Git terminology.

This workshop will explain you how Git works along with training you to use it on the command line. You will learn how the Git commands work and practice using them on many small exercises.

This will start with a set of simple common commands, and some simple concepts. Then over the course of the workshop, it will cover some more and more important sets of commands, and some more and more advanced concepts.

Notes :

It is best to come to this workshop with a laptop where Git is already installed.

This presentation will be the official kick-off of Sophia Antipolis The Thing Network Community. 

Deep learning recently led to phenomenal breakthroughs in a myriad of practical machine learning tasks, such as ImageNet classification challenge and playing the game Go. Yet many engineers and researchers are in quest of mathematical and theoretical understanding which aims to penetrate through the mysteries that explain clearly how it works so well in empirical success, especially to exploit open-source software and tune the deep neural network architecture. This tutorial will review recent work that aims to provide a mathematical justification for properties of special classes of deep networks, such as global optimality, invariance, and stability of the learned representations. The penultimate goal is to be able to understand the strength and drawbacks of existing open-source tools.

This talk is composed of a general introduction about DNN and three technical parts. Each part includes a formal discussion on a mathematical concept and some practical works with Python to be familiar with this concept. A 10-minute break will take place between part 1 and part 2.

1- Introduction (20 minutes) – This introductory talk will briefly review the recent success of deep architectures and the main open-source frameworks to run deep learning neural networks. It uses recent results to motivate the following theoretical questions: – How to deal with the challenge that the learning problem is non-convex? – What is the importance of “deep” and “convolutional” in CNN architectures? – What statistical properties of inputs are being captured/exploited by deep networks? – How can we add robustness to the learning of the network?

2- Part 1 – Global Optimality in Deep Learning (presentation 20 minutes + practical works 30 minutes): One of the challenges in training deep networks is that the associated optimization problem is non-convex and hence finding a good initialization would appear to be essential. Researchers have tackled this issue by using different ad-hoc or brute force initialization strategies, which often lead to very different local solutions for the network weights. Nonetheless, these local solutions give roughly the same results. This lecture will present a mathematical analysis that discusses the difficulty to find the global optimum in training deep networks.

3- Part 2 – Data Structure Based Theory for Deep Learning (presentation 20 minutes + practical works 30 minutes): One of the arguments given for the success of deep networks is that deeper architectures can better capture invariant properties of input objects. While a mathematical analysis of why this is the case remains elusive, recent progress has started to shed some light on this issue for certain sub-classes of deep networks. Scattering networks are a class of convolutional networks whose success lies on the ability to preserve discriminative information while generating stability with respect to high-dimensional deformations.

4- Part 3 – Generalization Error (presentation 20 minutes + practical works 30 minutes): This part will focus on data structure based theory for deep learning and discuss two recent developments. We first study the generalization error of deep neural network, including a discussion on fooling a deep neural network. Fooling a DDN consists in producing images totally unrecognizable to human eyes that DNNs classify as familiar objects. Next, we will discuss how deep learning can avoid the curse of dimensionality for certain class of functions. The curse of dimensionality means that the number of parameters in the neural networks can be exponentially large in certain cases.

The serverless is the new computer trend of recent years. With Microsoft Azure, which publishes a managed Kubernetes, known as Azure Kubernetes Service, and through open-source projects like OpenFaaaS or .NET Core, it’s now possible to start a resilient website with native https in minutes . During this conference, we will live code a real web site in .NET Core, exposed on the Internet with https natively, and supported by some functions, thanks to OpenFaaS, an open-source framework above Kubernetes.

The goal of the lab is to have people hands on kubernetes, manipulate core concepts such as pod, deployment, configmap, services (labels and selector mechanism), endpoints, job. From a pre installed application we will guide people th rough various use case, from application setup and rollout to investigation. This lab could be considered a learning by practice session. Attendees’ laptops will be used to connect to a prepared kubernetes instance.

The following core concepts will be described and manipulated: Deployment Pod Container configuration Volumes Probes ConfigMap Services Labels and Selector Endpoints Job.

Attendees will have to join with their own laptop and will have to connect to wifi. The lab will happen online on a preprovisioned kubernetes cluster.

The lab will be animated by David Benque and Cedric Lamoriniere.

Reinforcement learning (RL) is an area of machine learning concerned with how machines take actions in order to optimize a given reward by interacting with its dynamic environment. Some well-known recent applications include self-driven cars, and machines playing games better than humans (e.g., chess and go). One of the main advantages of this approach is that there is no need to explicitly model the nature of the interactions with the environment.  In this study, we present a new airline revenue management optimizer based on RL. The model does not need neither demand forecasting nor customer modelling to work. It is theoretically proven that RL will converge to the optimal solution. However, in practice, the system may require a lot of data (e.g., thousands of years of historical bookings) to learn the optimal policies. To overcome these issues, we present a novel model that integrates domain knowledge powered by a deep neural network trained in specialized hardware. The results show very encouraging results with different numerical /simulated scenarios. We believe this opens the door to a new generation of revenue management systems that could automatically learn by interacting with competitors and customers, so it can react much faster to market changes.

Notes :

In this study, we demonstrate the utility of RL in a real application such as RM optimization. For this purpose, we use an air travel market simulator in which an RL system and different RM systems are implemented. To the best of our knowledge, we are the first to propose a domain knowledge integration process to solve the warm up problem, allowing the deep neural network to initialize with the current policies of RM. This process is particularly important: it does not only provide a good starting point for RL, but also makes the training more stable and allows RL to adapt faster to market changes.

No particular technical requirements.

Elasticsearch that does the main job of analyzing and searching in large volumes of data.
Logstash as ETL to extract and enrich.
Beats as light collection agent.
Kibana as a visualization tool.
This workshop will give you an overview of the possibilities offered by these 4 tools, how they interact together and especially how they can help you solve your problems!

Notes:

The prerequisites are described on this page: https://xeraa.github.io/vagrant-elastic-stack/

It is absolutely necessary that the students have successfully completed the instructions on this page before they can join the workshop.

Cette présentation s’articulera autour des points suivants: – Introduction rapide des principes de la 4G – LTE , – Les canaux de communications, comment les objets “parlent” au réseau ? – Comment on optimise la portée et la consommation ? – quelques examples d’objets utilisant NB-IoT – Conclusion – ouverture vers la 5G

STAMP: Software Testing AMPlification for DevOps teams https://stamp.ow2.org/ =====

Overall Concept

The founding concept behind the STAMP project is that applying automatic transformation, a.k.a test amplification, to testing assets that are usually written by hand can greatly improve testing effectiveness. The STAMP project was launched to develop test amplification tools that increase levels of automation in software testing. STAMP focuses on test amplification in the context of DevOps and targets the early detection of regression bugs. The testing amplification tools developed by the STAMP project aim at helping DevOps to: – detect more regression bugs on continuous integration servers, before functional testing. – detect more scalability bugs, before going to production and experiencing bad behaviour (trashing, freezing) with high user load. – reproduce more production bugs in edge cases thanks to semantic logging.

STAMP Project

STAMP is an European project (Horizon 2020 program). The project gathers ten members including four academic partners with strong software testing expertise, an open source consortium (OW2) and five software companies in e-Health, Content Management, Smart Cities and Public Administration.

STAMP Open-Source Tools

The results of the STAMP project are open-source tools that are available on GitHub.
https://stamp.ow2.org/ For instance, – the DSPOT tool automatically generates new JUnit tests by modifying existing ones, – the Descartes tool verify your test suites to see if they really can detect possible bugs – the EvoCrash tool automatically crates non-regression test from exception appearing in your production log.

These tools are notably integrated in maven and jenkins pipelines to integrate easily your DevOps processus

STAMP Use Cases

Use cases provide strong experimental foundations to the STAMP project. They come from five different application domains (e-Health, Content Management, Smart Cities and Public Administration). The development of amplification technologies is carried out tight collaboration with the use cases so as to ensure the delivery of industry-relevant solutions.

In Activeeon, we leverage the STAMP tools to increase the quality of the open-source code of our products https://github.com/ow2-proactive. Notably, we improve our code coverage and we transform the events occurring in our production SaaS platform into new tests.