Coding Challenge #92: XOR Problem

References:

Videos:

Live Stream Archive:

Timestamps:
0:00 Introduction
1:15 Machine learning
3:00 Supervised learning
4:20 XOR problem
7:07 Hidden layer
11:49 Training data
16:26 Visualize results
19:30 Initializing weights
20:15 Learning rate
22:33 Add more hidden nodes
23:40 What's next?

Editing by Mathieu Blanchette
Animations by Jason Heglund
Music from Epidemic Sound

#neuralnetwork #machinelearning #gradientdescent #xorproblem #javascript #p5js

[NatetheAceOfficial]

A 25min coding challenge on my Monday morning commute! Great way to brighten the start of the week.

[RocketLR]

Mind = Blown! I need to figuer out ways to implement this in my work!

[grainfrizz]

Oh, Dan. You don't know how much you brighten up my day after seeing 16:15.

[Wyldone]

Maybe some simple observations, but it also shows two different behaviors.
'stuck' - analysis paralysis - can't decide between choices
'different pattern' - some people 'think differently' but there are several patterns that can come to the same conclutions.
'Faster learning' - some times more 'nodes' helps. Kind like two choices three (or more nodes) equals tie breaker or conclution checker

[AlexanderZotov]

Excuse me but are you building a Skynet?

[MrBa]

A good starter project for someone who just finished this series is the "Iris flower dataset". It contains data and answers for 150 flowers, with 3 different classes "categories". The data is nice and easy to read and use in whatever programming language you are using. The whole point is training your neural network to figure out what class of iris it is, based on some meassurements.

[crisiberny]

I don't know anything about neural networks but this was amazing! Keep the videos coming!

[yarilcaos]

Greetings from mexico!, please do more Neural Network stuff, thank for all your awesome job

[Video-Notes]

I just want to say Thanks to you Dan your way of teaching is incredible. Before comming here I dont had any idea of how can i implement dynamic chain rule differentiation and Now I am able to build my own multiple hidden layer neural network from scratch thank you very much. keep up the good work

[themirlabs]

i have been following along. At first i thought myn wasnt working but i gave it like 1000 training cycles on a bunch of randomly generated points and it started working!!!

I cant belive my MLA is working!

[CyberYeet]

I love your videos. I’m very glad I found your channel

[numeromojeangering]

There is and equation for XOR !! : abs(A - B)

abs(0 - 0) = 0
abs(0 - 1) = 1
abs(1 - 0) = 1
abs(1 - 1) = 0

[Seba_World]

I sub due to this guy enthusiasm. Need to follow all 10.x topic from this channel about neuron network. Spreading knowledge is smt will push our civilization to the next level (thinks like advancing our Civilization to type 1 according to Kardaszev scale etc). Cheers from Poland 🇵🇱

[veggiet]

The other thing that may or may not have helped is training it on a fractional input... Especially if you want to control the shape, but I wonder if it would help Keep itfrom getting stuck

[guinea_horn]

Why are there multiple correct solutions? Does it not really matter what's in the middle so long as the corners are the correct values?

[jeesi]

16:55 "I'm going to use 'i'"
I died a little inside

[raghavgarg]

Awesome Sir,
Please make more vedios on it

[hakanahlstrom]

Thank you for your efforts! Has the video about neural network digit prediction been released yet?

[CMRkiller]

I still don't understand the image being made. Is the color dependent on the neural network's certainty of what the output should be?

[SpicyMelonYT]

QUESTION!!!
how would i add data into the training_data variable with the format
{


inputs: [mouseX/width, mouseY/height];
outputs: ["0 or 1"(this part would be 0 for 50 clicks and 1 for 50 more)]


},


i tried doing this


training_data += ...


however this didn't work and i didn't really expect it to.
Do i need to use an array here instead to feed in the values from the clicks or maybe some sort of push() pop() thing?