Neural Networks: A Primer
summary
Explains how a perceptron becomes a multilayer network by training weights with backpropagation and stochastic gradient descent, using a concrete example: four inputs (speed, terrain, vision, mileage), five hidden neurons, sigmoid outputs and a predicted crash probability of ~0.925. Shows data prep with min-max scaling, loss as squared error with a 0.5 learning rate, and weight updates via gradient descent. Demonstrates Python stack (numpy, pandas, TensorFlow/Keras, scikit-learn, Jupyter) and a Movidius Neural Compute Stick producing sub-10ms inference on an ImageNet model.
AI-written summary · about
chapters (28)
- 0:00Start
- 1:54Stephen Hawking
- 5:19Artificial Intelligence
- 6:13Super Intelligence
- 7:29What Exactly Is Machine Learning
- 9:02Steps in Machine Learning
- 10:58Training a Model
- 11:16History of Neural Nets
- 12:41Back Propagation
- 16:07Perceptron
- 17:17An Activation Function
- 18:35Code a Perceptron
- 19:13Sigmoid Function
- 22:32Min Max Scaling
- 23:52Simple Min Max Scaling
- 28:25Weight Bias
- 31:01Data Types
- 33:33What Happens When We'Re Training a Neural Network
- 33:58Cost Function
- 35:28Learning Rate
- 38:33Gradient Descent
- 39:43Stochastic Gradient Descent
- 39:58Calculating these Weights
- 41:03Unsupervised Learning
- 44:04Stopping Condition
- 45:38Tensorflow
- 48:14Convolutional Neural Network
- 56:06Resources
related talks
01h 02mNeural Networks: A Primer
54mNeural networks by example
52mNeural Networks Walkthrough
40mNeural Networks: Where do I start?!?
27mDemystifying Deep Neural Networks
54mIntroduction to Understanding Deep Learning
46mFrom Perceptron to Deep Learning at Industrial Scale
43mA Gentle Introduction to Neural Networks