Mental Model
A NN can be thought of as an object that, builds squiggles and sequentially combines them into higher-level ones, until finally combining and sometimes squashing the final set of squiggles into surfaces that correctly fit the model’s training examples.
Symbols Key
Example
Given the following training examples with 2 input features and , and a corresponding output ; such that an example is a blue dot if its output is 0, and is a red cross otherwise.
The following surface seems to fit these training examples well.
Now, how do we construct an equation that represents that surface? it isn’t obvious to us how to do that, but the output layer does it by combining the high-level squiggles that it receives from the final hidden layer.
