Our keynote speakers are professors Marc LAVIELLLE and Leo LIBERTI of the École Polytechnique de Paris.
Directeur de Recherche, CNRS
Professeur, LIX, Ecole Polytechnique, France
2015 – IBM Faculty Award
Day: Monday, 27th of February
Optimization in the Real World.
How many times did you hear a CEO on the news say, “we optimized our strategy”? How many job offers you perused contained the word “optimization”? There are comparatively fewer positions advertising a need for “operations research” and “optimization methods”, though I’m betting you found some of those, too. On the other hand, plenty of job descriptions that appear to have nothing to do with optimization actually advertise positions that are mainly related to such techniques. Come discover what “optimization” means to different layers of society by means of this tutorial on Mathematical Programming — a formal language invented exactly for the purpose of describing optimization problems. The tutorial consists of an hour-long talk and two hours of hands-on tutorial where, if you bring your laptops, I will strive to teach you the ropes of this fascinating language.
Directeur de Recherche, Inria Saclay – Île-de-France
Head of joint Inria – CMAP team Xpop
Member of the Center of Applied Mathematics (CMAP), Ecole Polytechnique
Member of the High Council for Biotechnology
2015 Inria – Académie des Sciences – Dassault Systèmes Innovation Award
2015 ISoP (International Society of Pharmacometrics) Innovation Award
Day: Tuesday, 28th of February
Modeling and simulation of pharmacometric models: methods and tools.
Population models describe biological and physical phenomena observed in each of a set of individuals, and also the variability between individuals. This approach finds its place in domains like pharmacometrics when we need to quantitatively describe interactions between diseases, drugs and patients. This means developing models that take into account that different patients react differently to the same disease and the same drug. The population approach can be formulated in statistical terms using mixed effects models.
Such framework allows one to represent models for many different data types including continuous, categorical, count and time-to-event data. This opens the way for the use of quite generic methods for modeling these diverse data types.
In particular, the SAEM (Stochastic Approximation of EM) algorithm implemented in the Monolix software (http://lixoft.com/products/monolix) is extremely efficient for maximum likelihood estimation of population parameters, and has been proven to converge in quite general settings.
Monolix is associated with Mlxtran, a declarative language designed for encoding hierarchical models, including complex mixed effects models. Mlxtran is also a particularly powerful solution for encoding dynamical systems represented by a system of ordinary differential equations. Mlxtran is also used by mlxR, a R package for easily computing predictions and simulating data from complex mixed effects models (http://simulx.webpopix.org).
I will first show how these tools can be used for modelling and simulating pharmacokinetics and pharmacodynamics, infectious disease or tumor growth processes.
During the hands-on session, we will use the Monolix software for modelling pharmacokinetics data coming from a real word clinical study.
During the afternoon, conferences will be delivered by the following entrepreneurs.
Day: Monday, 27th of February
QuantifiCare (14:00 – 14:30)
Matilde GONZALEZ, Ph.D, Research Engineer
QuantifiCare 3D imaging solutions.
For more than two decades, QuantifiCare has been developing, manufacturing and marketing innovative imaging solutions for clinical trials and physicians. QuantifiCare is a global company specialized in 2D and 3D photography of the skin and 3D simulation of aesthetic surgery for breast and face. We offer compact and portable 3D photographic systems for standardized photo-documentation dedicated to dermatologists, plastic surgeons and aesthetic practitioners.
Altran (14:30 – 15:00)
Nicaise CHOUNGMO FOFACK, Ph.D., MSc, Ing
Short Bio: Graduated in June 2009, I got my Ph.D. degree in Computer Science from University of Nice Sophia Antipolis under the supervision of Philippe Nain and Sara Alouf on Feb. 2014. After a Research Scholar Visit of Prof. Don Towsley at University of Massachusetts, Amherst in USA, I held a Postdoctoral Research Fellowship at Orange Labs until Dec 2014. Then, I have been working as Data Scientist consultant at Orange Business. Late on Dec 2016, I joined ALTRAN TECHNOLOGIES as Architect Practice Data Science Lead.
Time series forecasting under hard constraints.
Data-driven business has gained an important attention in all-sized companies, since they can learn from their historical data what may happen in the near future. In this talk, we focus on Time Series Datasets which are very often used to describe a real-time activity. Precisely, we study the problem of “forecasting under hard constraints”, i.e. using minimal available information and part of data is missing. We derive an unsupervised machine learning framework and provide its implementation in a Maven project. Our solution requires no parameter tuning from the user.
Safran (15:30 – 16:00)
Short Bio: Alexandre Reiffers is a researcher at SafranTech where he is working on comparison of maintenance strategies. He received the B.Sc. degree in mathematics (2010) from the university of Marseille, the master degree in applied mathematics (2012) from the university of Pierre et Marie CURIE and the Ph.D. degree in computer science (January 2015) from the INRIA (National research institute in computer science and control) and the university of Avignon. His supervisors were Eitan Altman and Yezekael Hayel. His Ph.D is dedicated to the use of game theory for the understanding of online social networks. His areas of interest include convex optimization, stochastic processes, optimal control, routing games, bi-level problem and differential games; all applied to social networks, business comparison and internet economics.
Data indexation from the test bench with outlier detection
Safran is interested in the creation of tools for the indexation and analysis of test bench signals. Data are extracted from benches used in the conception and production phase. Indeed, when a failure occurs, we need to have a fast access to this data in order to able to know the origin of the failure.
Kitware (16:00 – 16:30)
Bastien JACQUET, Ph.D.
Open-source toolkits: the examples of 3D reconstruction MAP-Tk
Motion-imagery Aerial Photogrammetry Toolkit (MAP-Tk) is an open source C++ collection of libraries and tools for making measurements from aerial video. Initial capability focuses on estimating the camera flight trajectory and a sparse 3D point cloud of the scene. These products are jointly optimized via sparse bundle adjustment and are geo-localized if given additional control points or GPS metadata.
This project has similar goals as projects like Bundler and VisualSFM. However, the focus here in on efficiently processing aerial video rather than community photo collections. Special attention has been given to the case where the variation in depth of the 3D scene is small compared to distance to the camera. In these cases, planar homographies can be used to assist feature tracking, stabilize the video, and aid in solving loop closure problems.
The MAP-Tk GUI supports visualization and computation of depth maps, through state-of-the-art algorithms such as PlaneSweepLib (ETH Zurich).
MAP-Tk uses the KWIVER software architecture. Originally developed for MAP-Tk, KWIVER is highly modular and provides an algorithm abstraction layer that allows seamless interchange and run-time selection of algorithms from various other open source projects like OpenCV, VXL, Ceres Solver, and PROJ4. The core library and tools are light-weight with minimal dependencies (C++ standard library, KWIVER vital, and Eigen). The tools are written to depend only on the MAP-Tk and KWIVER vital libraries. Additional capabilities are provided by KWIVER arrows (plugin modules) that use third party libraries to implement various abstract algorithm interfaces defined in the KWIVER vital library. Earlier versions of MAP-Tk contained these core data structures, algorithms, and plugins, but these have since been moved to KWIVER for easier reuse across projects. What remains in this repository are the tools, scripts, and applications required to apply KWIVER algorithms to photogrammetry problems.
Koris International (16:30 – 17:00)
Asset Management: Improving Stock Market Volatility Estimates Using Trading Volumes
Volatility is an ever-present risk measure in the asset management industry. Whether volatility is used as a key input of the allocation program or only as a requirement for risk management, volatility estimates can bear considerable consequences for investment professionals. In order to disentangle transient from persistent volatility variations, we present a two-factor volatility model to study the impact of news arrival and trading volume on stock returns variance. The common observation that large volumes are associated with high volatility is explained by the fact that unexpected shocks in volume increase volatility, which is not the case for expected volumes of trading. Finally, we find that unexpected shocks in volume and the persistent component of the model are both main drivers of volatility dynamics.
Day: Tuesday, 28th of February
ExactCure (14:00 – 14:30)
ExactCure and personalized modeling of drugs: How to translate science into a company.
ExactCure is a newborn company that aims at modeling the precise action of a drug on a precise patient.
Taking the right drug for the right patient at the right time is a critical issue in modern health. There are more people dying of inappropriate medication than car accidents and suicides together! The cost has been estimated to 10 billion euros per year, only for the French Social Security.
Why is it so complex? We are all different; we differently respond to drugs, but we usually have one same posology for everyone. The principle of personalized medicine is to find the exact treatment for each patient. Now ExactCure opens the way to what we call personalized modeling.
In order to achieve this ambitious goal, ExactCure leverages an expertise in pharmacology, dynamic systems and control theory. This talk from Frédéric Dayan, ExactCure’s founder, will be focused on the transition between a scientific project and a real company.
GeometryFactory (14:30 – 15:00)
Invited speaker: Andreas Fabri
Title: CGAL – The Computational Geometry Algorithms Library
I will present the co-evolution of the CGAL Open Source project and GeometryFactory, the software development company that emanated from this project in 2003.
We start with the technical challenges encountered when developing geometric data structures and algorithms, and examples for how they are used in different application domains.
We then explain our strategy to turn research prototypes into products and make them perennial, the various forms of interaction between GeometryFactory and its academic partners, and the choice of an open source license that corresponds best to our business model.
Bentley Systems (15:30 – 16:00)
Modeling reality in 3D to advance the world’s infrastructure
I will present how some academic research work at the crossroads of computer vision and computational geometry has grown into ContextCapture, a high end software product automatically generating high resolution 3D models from simple photographs. This story starts in the labs, continues in the form of a thriving start-up, Acute3D, and succeeds in the adoption by Bentley Systems, a global leader developing software to better design, build and operate infrastructure.
Wever (16:00 – 16:30)
Dr. Brice EICHWALD
Wever, a new generation of carpooling
Wever is a community application that connects persons making travel on identical or partly identical routes. The promise of value that Wever made to users is that by combining their trips together they can optimize their routes (choice of the mode of travel, cost-sharing, real-time information, opportunities on the route).
Hitachi (16:30 – 17:00)
Deep Learning for Self-Driving Cars
Human factors appear in the vast majority of traffic accidents involving human injuries or deaths. Self-driving cars have been proposed as a possible solution to reduce traffic accidents, as well as improving commuting comfort, relieving traffic congestion and increasing fuel efficiency. In addition to the ethical and cybersecurity issues, numerous technical challenges arise in the design of autonomous vehicles. Among them, getting an efficient understanding of the scene and responding to unexpected events are vital to correctly planning trajectories.
Traditional approaches focused on explicit modeling of the environment and hand-crafted rules, but the last few years have seen the development of numerous extremely effective methods based on deep learning. Indeed, with the large number of available training datasets and the efficient processing power available on embedded systems, it has now become possible to train a deep neural network solely based on the recordings of the various sensors mounted on a car and use it on a real autonomous vehicle.
In the second part of the presentation we discuss current work on the desired properties on a deep neural network. Making sure that the autonomous systems are generic enough to account for the variety of driving environment and resilient to sensor dysfunctions can be handled directly in the training stage. Keeping the neural networks implementation efficient is a permanent struggle with the discrepancy between standard and embedded computing power: We describe how smaller neural networks can be generated from fully trained ones while keeping their desired properties. Although several autonomous systems focus on standard feedforward convolutional neural networks, recurrent neural networks have shown superior performance in steering angle and speed predictions, in part due to their ability to process temporally dependent events.
Building an autonomous navigation system is and will remain a challenge for the years to come, in particular if we want to understand it. We believe it will not only give us safer roads and vehicles, but also help us uncover the fundamentals of human cognition.
Wildmoka (17:00 – 17:30)
Thomas MENGUY, Co-Founder
Next generation video production
As a startup, Wildmoka is set to transform the way professionals produce videos for the digital age. Viewing habits, and thus dollars are dramatically shifting from linear TV to multi-form video consumption: The need to produce more content, adapted to different platforms like Facebook, Youtube, Snapchat, Instagram, etc is exploding.