Loading slide

Loading contents...

[█░░░░░░░░░░░░░░░░░][█░░░░░░░░░░░░░░░░░░░░░░░░░░░]1 / 33
<back>next

Something inside has to change

The last module ended with a different idea. Instead of writing every decision as a rule, people could show a machine examples with the correct answers attached.

But examples cannot teach anything unless something changes after a mistake.

A light switch has two fixed positions: off and on. A dimmer can move through all the positions between them.

Researchers needed something more like the dimmer. They needed adjustable parts inside the machine, parts that could move a little after each wrong answer.

For inspiration, some of them looked toward the brain. They borrowed one small idea from its connected cells and built a much simpler mathematical version: the artificial neuron.

Move the control. The important idea is not the light itself. It is that the setting can change.

# citations(3)↓
  1. [1]doi.org
  2. [2]deeplearningbook.org
  3. [3]ncbi.nlm.nih.gov