Create an MLP Neural Network from Scratch (NumPy and MNIST only)
Do you really want to understand how a neural network works? In this video, we build an MLP from scratch using NumPy to recognize MNIST digits, without frameworks (no PyTorch, Keras, or TensorFlow). If you want to truly understand forward and backprop logic, we'll program it by hand.
You'll see step-by-step:
✅ Data preparation (MNIST)
✅ Activation functions
✅ Forward pass (prediction calculation)
✅ Loss function
✅ Backpropagation (manual gradients)
✅ Gradient descent and training
✅ Evaluation and prediction examples
📌 Video repository/code: https://github.com/hugoramallo/DeepLe...
📌 Colab repository: https://colab.research.google.com/dri...
📌 Dataset (MNIST): https://www.kaggle.com/datasets/hojja...
🎓 AI Courses and Resources: https://www.globalcobots.com/
🔔 Subscribe: https://www.youtube.com/@globalcobots...
CHAPTERS:
00:00 - Introduction and Objective
00:38 - Network Architecture and Mathematics
04:08 - Creating the NeuralNetwork Class (Init)
06:43 - Implementing the Forward Pass
08:11 - Implementing Backpropagation (The Hard Part)
10:50 - Preparing the MNIST Dataset
12:47 - Training Loop
16:38 - Model Evaluation (Accuracy)
17:45 - Visual Predictions (Inference)
18:40 - Conclusions and Closing
Cylinder Five by Chris Zabriskie is licensed under a Creative Commons Attribution 4.0 International License. https://creativecommons.org/licenses/...
Source: http://chriszabriskie.com/cylinders/
Artist: http://chriszabriskie.com/