Vision, image and machine learning (partie AM) - TP

Tutorial (TP) #

The aim of this tutorial is to implement from scratch a simple neural network trained by backpropagation. This will be a formative introduction to more advanced frameworks (such as PyTorch) where learning is automated.

The equations and small code (1D) presented in the course

The two exercises (all steps are described inside the code.):

• Simple 2D classifier, everything in one loop
• MNIST classifier with classes

Documents to read #

• The equations behind backpropagation with code and explanations
• The backpropagation equations
• Introduction-au-deep-learning(Miximum)

Solutions #

• Correction 2D classifier from scratch
• Correction MNIST classifier from scratch

Variant

• From Miximum