 Hello everyone! My name is Dennis. I'm an application engineer at ST Microelectronics. Today I'm going to show you how to design a machine learning solution for a real application step by step. So, make yourself comfortable and let's get started. The pencil motion detection application we are going to analyze is intended to provide an example where power consumption is a critical aspect due to the form factor. Our LSM6DSOX fixed on a pencil will be programmed in order to detect these four different scenarios or classes. Steady, pen is on the table. Idol, pen is in your hand but you are not writing. Writing, pen is in your hand and you are actually writing. Or other, not covered by the other cases. This slide is meant to summarize a high level comparison between a traditional approach versus a machine learning approach. In particular, the one available with LSM6DSOX. As you can see, the main difference here is related to the offline data analysis. With a traditional approach, good understanding of signal processing, optimization techniques and the usage of different programming languages are needed for an algorithm implementation. Moreover, new data set will require to reiterate the expensive process. In contrast, machine learning approach is based on a set of tools that will reduce your effort to almost none. This is just an example that shows few key parameters at system level using LSM6DSOX. On the left side, logic is completely running on a microcontroller or SOC. While on the right side, all the logic is executed inside the sensor. As you can immediately observe, power and coding efforts are key factors in a machine learning approach. Those are the main steps needed in order to realize a fully working solution based on machine learning. First step is the data collection, and we are going to do that using Profimamps board together with Unico GUI. After data have been collected and labeled, we will extract features, and if needed, apply filters. Unico GUI will help us during this step. Features are basically the input for a machine learning algorithm. With an external tool such as Weka, we will be able to train and generate a decision tree. The decision tree will be then loaded back into the Unico GUI in order to generate the register's configuration needed to run the decision tree into LSM6DSOX.