What the hidden layer learns

The network learns by fixing mistakes. But what exactly does it learn?

Before that: for any of this to work, every training example needs a correct answer attached to it.

The network makes a prediction, compares it to the right answer, and measures how far off it was. That gap is what drives the whole process.

If you're training a network to tell cats from dogs, someone had to go through thousands of photographs and label each one, "cat," "dog," "cat," "dog," before training could even begin.

The network never figures out the right answer on its own. It just learns to get closer to the answers it was given.

What the hidden layer learns

The network learns by fixing mistakes. But what exactly does it learn?

Before that: for any of this to work, every training example needs a correct answer attached to it.

The network makes a prediction, compares it to the right answer, and measures how far off it was. That gap is what drives the whole process.

If you're training a network to tell cats from dogs, someone had to go through thousands of photographs and label each one, "cat," "dog," "cat," "dog," before training could even begin.

The network never figures out the right answer on its own. It just learns to get closer to the answers it was given.