دانلود کتاب یادگیری ماشین توکار با میکروکنترلرها – کاربردها در بردهای توسعه STM32

Our brain is continuously exposed to data originating from our sensory organs as the eye, nose, ear, skin, and tongue. In other words, each sensory organ converts the world around us to data. Then, the brain performs two fundamental operations on this data: First, it learns what information the provided data is offering. Second, it applies inference based on the learned information. These operations are more or less the same for most living organisms. The aim here is interacting with the outer world in an intelligent manner. We can also form intelligent systems interacting with the outside world. To do so, the system should gather data from its surrounding through its sensors. There should be a dedicated unit on the system to execute a learning algorithm and then apply inference to new data. As a result, the system can learn from data acquired from the environment and perform inference accordingly. Machine learning methods are introduced for this purpose. We can group existing machine learning methods into two parts as traditional and neural network-based. Traditional machine learning methods are based on designing or forming a system via given data. On the other hand, neural network-based methods directly learn from the data. One approach may be more useful compared to the other for the problem at hand. Therefore, we will cover both methods in this book. There are several ways in which machine learning methods can be used in an intelligent system. We can group this usage into three main groups as classification, regression, and clustering. The classifier assigns the incoming data to one of the known classes. Regression leads to predicting future values based on given data. Clustering groups the data based on its inherent characteristics. As a result, all these machine learning methods lead the system to interact with the outside world in an intelligent manner. Recent machine learning methods (especially neural network-based ones) require vast amount of data and huge computation power during training. Hence, a very powerful PC or workstation may be needed for operation. While doing so, data is collected from sensors. Then, it is stored in a local device or on cloud. Afterward, the collected data is used in training. Depending on the selected model and its parameters, this training operation can be done in minutes to days (depending on the computation power of the used device). The size of the trained model after training can be from few hundred kilobytes to hundred megabytes.