The network view of the world

So what exactly is going on here, what is it that the network is identifying? Essentially, we are flattening our 3D view of the world into a 2D line or curve. A typical example of how this line may look normalized is as follows:

Normalized input points

Those inputs represent the normalized view the neural network is training for, or perhaps, against. If you trained the network to recognize that line, then the warning sound should go off when it detects the said line. Of course, the more points you add, the better your recognizer may or may not work. We will leave it up to you to further test the network on your own.

Neural networks were quite popular with game and graphic developers in the late 1990s and early 2000s. NNs showed some success in various AI scenarios, driving games especially, but at the end, other purpose-built techniques won out, that is, until quite recently with the advent of new techniques such as convolutional NNs. These new successes have led to massive surges in deep learning techniques and platforms.

This simple NN can be extended to recognize other simple functions or patterns you wanted. However, it will work poorly if we try to use it for any of the other recognition tasks we identified earlier as critical for AR. Therefore, in the next section, we will look at how ML solves our recognition problems with a new platform developed by Google, called TensorFlow.

Table of Contents for The network view of the world

Create new playlist

Sign In

Sign Up

Table of Contents for
The network view of the world