Deep Learning In Hunt For Dark Matter, Better Crops And E-commerce Service

## Schedule ####

17:40 Open doors

18:00 Pawel Rosikiewicz, UNIL
Welcome Note

18.05 Patrick Barbey, Managing Director, Innovaud (Event Sponsor)
Startup Support in Vaud

18.20 Fabio Capela PhD, Data Scientist Plair SA
Wide neural networks for image classification

18.50 Jean-Baptiste Cordonnier & Brune Bastide, EPFL
Swiss avalanches: visualizing historical data for social good

19.05 Break

19.15 David Francois Droz, UnivGe
Searching for dark matter with machine learning

19.40 Luca Baldassarre PhD, Head of Data Science, Gamaya
Artificial Environmental Intelligence with High Flying, Far Walking and Deep Learning

20.10 Apero

## ABSTRACTS ##

Startup Support in Vaud

Region Vaud in Switzerland is known for its "deep tech" innovations in many areas from engineering to life sciences, from cleantech to digital innovations. There are a number of reasons for this dynamism, in particular the research institutions and a diversified talent pool. Key is also the support tools offered at the cantonal and federal levels. Support ranges from financial incentives and interest-free loans to free coaching by experts. A few examples of supported startups will be presented to illustrate the various help mechanisms.

Wide neural networks for image classification

Deep residual networks are among the most widely used machine learning algorithms for image recognition. However, adding extra layers to improve the accuracy of the system turns out to be expensive in terms of training time. In this talk, we are going to explain what are ResNets and how to circumvent their slow training process by decreasing their depth and increasing their width.

Swiss avalanches: visualizing historical data for social good

Risk zero does not exist in alpinism. Statistical models have been developed to assess this risk but they do not prevent tragedies. We do not claim that we can do better, but given that most of the accidents are due to people's bad decisions, we are convinced that raising concern about the past mountaineering accidents can strongly improve alpinists’ judgement in the future. The aim of this project is to gather meteorological and environmental data (weather condition, precipitations, snowpack, slopes, exposures…) along with avalanche cases and casualties. By leveraging means of interactive visualization, we provide the skiers ways to understand the conditions of previous cases and maybe hints that could have changed the outcome.

Searching for dark matter with machine learning

The mechanisms of acceleration and propagation of cosmic rays, and the nature of dark matter, are still open questions in astrophysics despite generations of ground and space detectors. The growing complexity of such experiments require physicists to design and deploy new analysis methods, among them machine learning. Recently neural networks started appearing seldomly in some high energy physics experiments, notably the Large Hadron Collider at CERN, but have not yet been applied to space-based detectors. In my talk I will present such an experiment, the DAMPE satellite of the Chinese Academy of Sciences. I will show how deep neural networks can significantly improve the accuracy of existing classification algorithms, despite the multiple constraints specific to particle physics and space borne detectors.

Artificial Environmental Intelligence with High Flying, Far Walking and Deep Learning

In order to meet a growing demand and reduced environmental resources, in the next few years agricultural businesses need to increase productivity, efficiency and sustainability. At Gamaya, we combine airborne hyperspectral imagery with satellite scenes and historical data, leveraging crop models and deep learning, to provide immediately actionable insights, for more efficient and sustainable farming. In this talk, I will delineate the information and data challenges of precision agriculture, illustrate our data science approach and provide examples of our analytical solutions.