Mathematical Engineering is the center of Vekia’s expertise. Our technical team is led by senior researchers in Machine Learning, coming from the high level academic research (National Center of Scientific Research - CNRS and National Institute for Research in Computer and Control Sciences - INRIA). They are assisted by senior developers in scientific programming.
The Machine Learning
The Machine Learning is a recent discipline resulting from the convergence of mathematics and computer science. These methods are currently widely used in biology, medicine, physics and finance. They can combine refinement of modeling and control of processing time. Vekia is the leading company that offers this technology for the distribution sector.
The founding principle of Machine Learning is to find the best match between the complexity of mathematical models and those of data to be processed. Before Machine Learning, modelers realized this match by test and error. The Machine Learning allows a systematic and efficient best model research. This ensures that our methods are both simple to use and robustness.
An example: predicting daily sales of fresh produce
Fresh products sales are influenced by various factors, which act more or less strongly depending on the product. For instance, fruits sales are strongly influenced by the season, while pastries sales are driven by scholar holidays. Many other factors may also occure, such as weather, prices, day of the week or month-end. To predict sales, we could take into account all these factors for all products. This leads to a complicated model that gives a result for all references. In return, its performance is poor: it is unable to balance, prioritize properly the impact of different parameters, and its predictive accuracy is often reduced.
Machine Learning allows choosing for each product what the factors are really influencing. It's completely transparent for the user! The illustration below shows a too complex model (left) "glue" fully data during the learning phase (the first two weeks), but it gave totally false figures in prediction phase. On the opposite, a too simple model (right) is unable to reproduce the impact of various factors. The right model (below) is automatically determined by our algorithm, Machine Learning: it works well during training and during prediction.
Mathematical Engineering for the service of your organization
Vekia has designed Provisia and Affluencia and with the Machine Learning principles, guaranteeing the best performance in sales forecasting, supply and crowds management. This is also specific studies and developments, made especially for you.