The Open Coffee Club organized a conference on Thursday June 2 on Machine Learning applied to startup projects.
And all this in an inspiring place, since we were received in the premises of CRITEO, world leader in digital marketing and innovator in the field of Machine Learning algorithms (or machine learning).
eiver was invited as a Startup using machine learning processes in the field of driver assistance and shared the microphone with 3 experts, Christel BELTRAN, Partner Executive at IBM and Watson Ambassador, and of whom SaveCode / eiver is partner, Alexis BISMUTH, Doctor of Statistics, Mathematician-researcher in the field of artificial intelligence and Franck BARDOL, Arts et Métiers, Data Scientist, Founder of the Machine Learning Group.
Demystification
A conference moderated by Eric DUBOIS and a very oriented “demystification” exchange aimed at the creators of technological, web or mobile projects, having to implement self-learning mechanisms or information corrections based on data acquisition in mass (Big data). We were able to present our machine learning experience by presenting our approach to acquiring and processing automotive usage data, based on capturing the instantaneous speed of the vehicle and translating this data into 3 “instantaneous states”: acceleration, constant speed, deceleration. This “splitting” of the trip data has led us to reclassify any portion of the trip into “patterns” or “canvas” of 3 types: Coaster, Cruiser or Liner, which, combined together, then make up any type of automobile trip. It is to this first step of transforming basic data (the Data), which is often unstructured and not analyzed, that we apply our first journey qualification calculation, which constitutes a 1st scoring of the quality of the journey. The less accelerations and decelerations we register, combined with the average speed of the portion of the journey and the distance traveled, the more the “quality” of the journey will be valued. The Scale is simple. On a weighting scale 0 to 2, a journey of average quality will be “scored” at 1, a journey with too many acceleration / deceleration per km will be scored less than 1 and journeys with few speed differences will be scored higher to 1.
Compensation mechanisms
And this is where the problem comes in: a driver called “urban” or “coaster” risks being disadvantaged on his score because he is constrained by the infrastructure on short distance journeys of medium low speed and high. stresses in speed differences. It should be noted here that this discrimination only concerns professional populations within the framework of their mission. eiver is not intended to encourage car journeys in town, if the individual has the choice of his mode of transport. But to meet the demands of our business customers, who implement our solutions as part of the management of their fleet of vehicles, we implement Machine Learning mechanisms that compensate for a constrained journey and regularly restore the balance on populations of professional drivers.
Machine Learning applied to obtaining the truth.
A second algorithm, applied to all drivers this time, makes it possible to refine the fuel consumption and CO2 emission values, based on the results already recorded by the community. With nearly 45,000 downloads of the eiver application in Europe and 4 million km traveled to date by our drivers, on any type and brand of light vehicle, eiver is able to refine the actual consumption values and apply them in according to the driving score mentioned above.
The implementation of machine learning mechanisms is now facilitated by several factors: the fluidity of mass data transfers, the multiplication of on-board sensors and smartphones, and above all their calculation capacity. But also access to offers and services in the cloud such as BlueMix or Watson to evoke the IBM solutions that we use and which benefit from excellent scalability, scalability and robustness, and above all data security.