Before writing our first program using TensorFlow, we will briefly go through the concepts that will help us to understand how the TensorFlow Lite model work. We won't be going into the details, but a conceptual high level overview alone for better understanding.

MobileNet and Inception V3 are the built-in models that are based on convolutional neural networks (CNNs). 

At its most basic level, CNN can be thought of as a kind of neural network that uses many identical copies of the same neuron. This allows the network to have lots of neurons and express computationally large models while keeping the number of actual parameters – the values describing how neurons behave – that need to be learned fairly low.

This concept can be understood with the analogy of a Jigsaw puzzle and how we usually solve one. The following diagram is a puzzle that needs to be solved: 

If we have to assemble this puzzle from the pieces provided, just think about how you will start solving it. You may group all the pieces with different colors together. Then within the same color, you'd check for patterns and then assemble them. This is the same way that convolutional networks train for image classification and recognition. Hence there is only a small portion, each neuron remembers. But the parent neuron understands how the things within its scope needs to be assembled to get the big picture.

In the Inception V3 and the MobileNet models, both work based on the CNN concept. The model is pretty much trained and stable. All we need to do to use our set of images is retrain the model with our images. So now that we have had enough of concepts and theory, we will move on to writing our first sample program using TensorFlow Lite for Android.

We will be using the TensorFlow for mobile for a classification application in Chapter 9, Neural Networks on Mobile