Issue link: https://resources.mouser.com/i/1481902
STMicroelectronics 2022 31 Deductive algorithms are based on the application of a general hypothesis that can predict with high precision what the observation of the behavior or object of interest will be. It is critical to have a robust and fully validated hypothesis in order to support this type of algorithm. On the contrary, inductive algorithms are based on generalization from specific observations of the behavior or object of interest. Inductive algorithms deliver more reliable predictions when larger amounts of collected data are available. In particular, the highest possible number of different cases is important for a robust outcome. Figure 5: A high-level description of MLC implementation in an intelligent sensor. (Source: STMicroelectronics) Figure 5 breaks down the MLC implementation into steps. The initial step is to determine what sensors the target application needs. The second step is to label the data using the features embedded in the sensor to identify what computational blocks to implement. Then, you need to build the decision trees. Step four is the implementation of the decision trees. In step five, the system collects new real-time data to test the decision trees. During the final step, the results are observed to evaluate the quality of the trained model. UART,... Gateway, ... Data Collection From Sensor 1 : Sensor n Process New Data Real time test Label Data • Filters • Features •Connections Outputs: Result 1: Result m Build Decision Tree • Classification • Results Embed Decision Tree • Decision Tree implementation Figure 5 Figure 4 Data Collection Define Outputs: Result 1: Result m Finite State Configuration 2. Instructions: • Conditions • Commands • Parameters 1. Data Section: • Fixed data • Variable data from Sensor 1: Sensor n " It is critical to have a robust and fully validated hypothesis in order to support this type of algorithm. Figure 4: Finite State Machine implementation in an intelligent Sensor. (Source: STMicroelectronics) The implementation of an FSM can be done in three steps. The first step is to define the inputs. That means identifying the sensor data to be used by the FSM. The second step is to define the states by means of commands, conditions, and parameters. The last step will be to define the output of the FSMs. Figure 4 shows this three-step FSM approach.