Programming involves explicitly telling a computer step-by-step instructions to complete a task, like making a peanut butter and jelly sandwich. Deep learning is different - it involves feeding large amounts of training data and desired outcomes to a system, allowing it to learn on its own how to achieve goals or analyze inputs, like being able to make a sandwich from any ingredients or recognize images. This shift from explicit programming to implicit learning through large datasets allows systems to complete tasks and handle unexpected inputs without needing to be directly programmed for every possibility.
4. Whoops!
New Step 1: Open the top of peanut butter by twisting
counter clockwise
5. Good Programming Bad Programming
In traditional programming, the computer isnt responsible for
the outcome. It just does what a human told it to do.
7. Good Sandwich Bad Sandwich
Deep Learning Starts From The End.
You label the outcomes. And you give it A LOT of data.
8. 1st Type of Deep Learning:
Understanding What It Sees (what Google Photos does)
NEW IMAGE
LOOKS AT BILLIONS OF
DETAILS FROM TRAINING DATA
(NOT HUMAN READABLE)
DECIDES:
NO TOP BREAD
WHEAT BREAD
JELLY > PEANUT
BUTTER
CRUST NOT REMOVED
9. Try peanut butter and
jelly on same side
2nd Type Of Deep Learning:
Trying To Achieve A Goal (what AlphaGo does)
Try
from the edge, from the center, wheat
bread, white bread, chunky peanut butter,
and millions of other ideas
Try peanut butter on
one side, jelly on other
10. The real magic is not this at the end
(although delicious it might be):
The magic is that even if you give it
unexpected inputs
11. It can still make this and this and this
And no human had to tell it what to do.
13. You want products to do what YOU expect.
Cars that dont kill people
Medical devices that dont miss a disease
Educational software that adapts to the child
Kitchen appliances that cook food thoroughly