Seasonal work due to COVID 19.

In this internship, I joined a team in the Research & Technology department at Safran Electronics & Defense. This team, made up of researchers and PhD students, aims to innovate in the field of automation, namely to control dynamic systems.The goal of my internship was to explore a new way to train neural networks so that complex constraints could be applied on the parameters of this neural network while its training.One of the steps to controlling a dynamic system is to mathematically model it. And neural networks allow us to have a model for complex dynamic systems.The topic of my internship was therefore to find out if it was possible to use optimization algorithms (of the genetic type) to train a neural network in order to modelize a dynamic system. This way of training makes it possible to apply constraints on the neural network and thus ensure that the model retains the characteristics of the dynamic system (for example, the stability).There was a part of searching for different algorithms to use for this and select the ones that returns the best training. And a second part of searching for the architecture of the neural network to use.The results of this internship are that training a neural network with genetic algorithms is not as satisfying as using Levenberg-Marquardt alogrithm. But, there was hope at the end of my internship, to explore the possible architecture, but I did not have enough time to dig deep into this part.

During this internship, I joined the R&D team of a start-up called Angus.ai. This start-up offers large-scale stores solutions to analyse the customer journey in stores, or even the life of the product on the shelves.The start-up focused mainly on the second solution, and aimed to provide store staff with assistance in order to streamline the re-sale of products.My role in the R&D team was to develop Computer Vision solutions to determine whether or not a product is present in the store shelves. The ultimate goal is to provide alerts to store staff in order to put a product off the shelf, or to re-present the product so that it is more visible to store users.I implemented some solutions for this problem. One of the solution has been put into production.