Sei sulla pagina 1di 48

Artificial

Intelligence,
what?
Machine Learning for Kids
Best is trying out yourselves.
Let’sQuickDraw
jump right into it!
quickdraw.withgoogle.com
DATA: doodle drawings, millions of them! https://quickdraw.withgoogle.com/data

JOB: guess what thing the doodle drawing is


What else can we do with doodle
drawingsAutoDraw
data? autodraw.com
DATA: doodle drawings + nicely drawn drawings

JOB:

1) guess what thing the doodle drawing is, THEN,


2) give nice drawings for it
WHY IS
THAT
CALLED
MACHINE
LEARNING
?
IMAGINE MACHINE =
POKEMON,
Train your pokemon! YOU = POKEMON
TRAINER Train with battles
Try different attacks

Result: win battles

Train your machine!

Train with data


Try different answers

Result: do jobs
Machine Learning, in SIMPLE
view Let’s get the machine to learn stuff,
by training it thousands times, millions times, billions times

DATA:

Training Data

...after it finished learning from the intensive training...

JOB:

Input Output / Action


Meh, I thought that’s obvious for
computers
Not really. Computers are actually stupid, oops,

I mean: has no intelligence.

Computers work like this:

Do exactly as I say Yes Sir!


Do exactly as i say
Me: Hey computer

Computer: yes

Me: do this for me:

1) Go to the shop
2) Get 3 chocolates
3) If they have candy
4) Get 12

Computer: sure!
Typically computers are
programmed like this
On the other hands, humans
have intelligence
Humans learn.
After consistent & persistent
learning & training

Learner becomes Master


What if computer is able to
learn?
We call it...
Can i try to
train
machine to
learn?
LEt’s train our pokemon, oops,
machine! Teachable Machine
teachablemachine.withgoogle.com
DATA: photos of objects, you will give these!

JOB: guess what object in front of camera


WHAT IF WE’VE TRAINED WITH
MORE PHOTOS
Giorgio Cam OF STUFF?
experiments.withgoogle.com/ai/giorgio-
cam/view/
DATA: photos of objects, already trained!

JOB: guess what object in front of camera


Ways to
learn
Supervised Learning

Apple

Banana

???
Unsupervised Learning

1 2 3 4 5

Hmm.. looks like 1, 3, 5 are in one group, while 2 and 4 are in another group
Reinforcement Learning

Okay, must get the ball but shouldn’t do it that way


Is that all?
Computer machine can learn to
play Flappy Birds!
Even the most difficult game, AI
have won
Including DOTA!
MORE FUN STUFF!
Flappy Bird : https://www.youtube.com/watch?v=xM62SpKAZHU (mins 3, 7, 2:23)

Tetris : https://www.youtube.com/watch?v=QBjtzRDDjys

MarI/O : https://www.youtube.com/watch?v=qv6UVOQ0F44

Games: https://blog.openai.com/reinforcement-learning-with-prediction-based-
rewards/
Practical
Application
s of AI/ML
Can you guess:
how did the
machine learn?
Use this

What’s the training data? What’s the results?

Training Data: Results:

.... ....
Autocomplete
Autocomplete

What’s the training data? What’s the results?

Training: Results:

Texts from what most people type Prediction of the most likely:

Texts from what the user him/herself - What’s the next letter
usually types - What’s the next word
- What’s the intended word
Search image by text
Search image by text

What’s the training data? What’s the results?

Training: Results:

Images with the labels Given the image, predict what’s the
labels
For example, images of cats are
labeled with “cats”, images of kids Automatically label them
playing on playground are labeled with
“kids”, “play”, “playground” Search the images using label
Face Recognition
Face Recognition

What’s the training data? What’s the results?

Training: Results:

Photos of faces from different angles, Predict the person’s name


labeled with the person’s name
Translation
Translation

What’s the training data? What’s the results?

Training: Results:

Texts mapped from language A to Predict the most likely translation


language B
Scene Recognition

http://cs.stanford.edu/people/karpat
hy/densecap/
NeuralTalk and Walk :
https://vimeo.com/146492001
Scene Recognition

What’s the training data? What’s the results?

Training: Results:

Photos of scenes with labels on the Predict what’s the likely scene
objects
Reverse it: Find scene

http://cs.stanford.edu/people/karpathy/deepimagesent/rankingdemo/
Voice-based Assistant
Android : OK Google, iPhone : Siri
Voice-based Assistant

What’s the training data? What’s the results?

Training: Results:

Voice recordings of people, labeled Predict the text


with the text
Predict and do the instructions
Text and the intended instruction
Self Driving Cars
Self Driving Cars

What’s the training data? What’s the results?

Training: Results:

Sensors, Camera stream data How to drive safely


+
Driving response
Autonomous Robots
Autonomous Robots

What’s the training data? What’s the results?

Training: Results:

Repetitive physical task Do the task


That’s all folks!

ainunnajib@gmail.com
Twitter: @ainunnajib
fb.com/ainunnajib
IG: @ainunnajib.id

Potrebbero piacerti anche