Paris Machine Learning #10 Ending Season 4, Large-scale video classification, community detection, Code Mining, Maps, Load Monitoring and.Cognitive.

At Qucit we use geographic data collected from hundreds of cities on a daily basis. It is collected from many different sources, often from each city's open data website and used as inputs to our models. We can then predict parking times, bikeshare stations occupations, stress levels, parking fraud… Gathering it is a lot of fastidious work and we aim at automating it. In order to do so we want to get our data from a source available everywhere: satellite images. We now have a good enough precision to detect roads, buildings and we believe that single trees can be detected too. We tested our model on the SpaceNet images and labels, acquired thanks to the Spacenet Challenge. During this challenge the Topcoder Community had to develop automated methods for extracting building footprints from high-resolution satellite imagery. Non Intrusive Load Monitoring is the field of electrical consumption disaggregation within a building thus enabling people to increase their energy efficiency and reduce both energy demand and electricity costs. We will present to you this active research field and what are the learning challenges @Smart Impulse.