Coding a 3-Layer Neural Network From Scratch in Python

We start by considering how to visualise classification scenario involving two classes of points on a two-dimensional plane. In the last lesson, we learned how to implement binary classification using a single artificial neuron. This time, we expose the limitations of using a single neuron and instead construct a network of neurons distributed into three layers. To implement such a network requires us to master computing long derivative chains. Luckily, by using the chain rule, and adequately visualising the process, this task becomes much less daunting. After computing the partial derivatives of the network's output in terms of all the weights and biases, we implement the neural network from scratch in the python programming language and then plot the classification results.

Lecturer: Dr. Nikola Marinčić, Chair for Digital Architectonics
23, March 2020, ETH Zurich

[xxh]

Thank you sir. Really appreciate you and your team are working on this remote lecture and publish online in this such period. I really enjoy the knowledge. Sincerely thanks.

[karimelfaramawy]

Brilliant course, can't thank you enough for this precious content!

[IasonGiraud]

really good material. very well explained for architects

[mar]

Please, provide the link to the notebook

[samuelbsua]

Thanks for the lesson, really happy to listen to this course

[artemgilmanov]

At 8:42 you say: "multiply point's coordinates with weight and biases and determine their class". Does it mean that you used all the cell's centres and applied neurone to find the boundaries? Then, did you assign a label for each point as well?