Tariq Rashid - A Gentle Introduction to Neural Networks and making your own with Python

PyData London 2016

Neural networks are not only a powerful data science tool, they're at the heart of recent breakthroughs in deep learning and artificial intelligence.

This talk, designed for a complete beginners and anyone non-technical, will introduce the history and ideas behind neural networks. You won't need anything more than basic school maths. We'll gently build our own neural network in Python too.

Ideas: - search for intelligence machines, what's easy for us not easy for computers.

DIY: - MINST dataset - simple network 3 layer - matrix multiplication to aid calculations - preprocessing, priming weights - 95% accuracy with very simple code! - improvements lead to 98% accuracy!

00:10 Help us add time stamps or captions to this video! See the description for details.

[sundareswaransenthilvel]

Thanks for the simplistic explanation of how neural network functions.

[ravisawhney]

Very nice intro to NN for non-programmers.

[invinity]

awesome talk! i finally understand how to build neural network in Python!

[KayYesYouTuber]

Hi Tariq, This is simply excellent. I also read your book now. This is very good.
I have a question. In the example you have used here, you have shown a classification problem. DO you have an example for a regression problem where I have some continuous X variables and some Boolean X variables, predicting a continuous Y variable.

[avonstar]

At 10:02 Tariq says "..its short and long thats a ladybird" Of course he meant caterpillar. Sorry to be so pedantic. The video is excellent. I learnt a lot.

[kloro]

On page 48 of Make Your Own Neural Network in the diagram of the NN, why does neuron 2 in layer 1 connect with all 3 neurons in layer 2 and the other neurons in layer 1 connect with only 2 neurons in layer 2? He uses the same diagram in his presentation.

[mariomueller]

Hello Tariq thank you very much for your really good book
I already bought it and studied. But there is something that I don't understand.

And I think it is a bug? Hopefully I am wrong and you can give me the correct answer.

I thought the sigmoid function is used in the back propagation.
But when I have a look in the slides (video 40:59) the the sigmoid is processed during the inbound/forward propagation (from the input-layer to the output-layer).
Can this be?

I appreciate your answer.
thanks regards, Mario

[Jables]

What type of neural network did we just watch being made?