Designing and training the ML model
With the dataset in our hands, we can start designing the model.
In this recipe, we will develop the following architecture with Edge Impulse:
Figure 6.17 – Fully connected neural network to train
As you can see, the spectral features are the input for the model, which consists of just two fully connected layers.
Getting ready
In this recipe, we want to explain why the tiny network shown in the preceding diagram recognizes gestures from accelerometer data.
When developing deep neural network architectures, we commonly feed the model with raw data to leave the network to learn how to extract the features automatically.
This approach proved to be effective and incredibly accurate in various applications, such as image classification. However, there are some applications where hand-crafted engineering features offer similar accuracy results to deep learning and help reduce the architecture's complexity...
