Deep Neural Network Intuition

Neural networks explained simply: why they were invented when equations ran out, then what a model’s size really means: parameters, bits, memory, training versus inference and fine-tuning.

Why this course

The sequel to Machine Learning Intuition. Why neural networks were invented, built up from the problems equations couldn't solve. Then what a model's size really means: parameters, the bits each one is stored in, memory, training versus inference, and fine-tuning. Then a quick peek inside one neuron: multiply, add, add the bias, and the ReLU switch that lets a network bend. Then how a network picks one answer out of many with softmax, how machines learned to see with convolutions, how words become numbers with tokens and a bag of words, how recurrent networks and LSTMs read words in order, and the big idea behind today's language models: predict the next word.

What you will learn

  • Why neural networks were invented
  • Neurons, weights and layers
  • Parameters and model size
  • Training versus inference
  • Fine-tuning

Course outline

8 video lessons, each starting from a problem, with quizzes and hands-on challenges in between.

  1. When Equations Aren't Enough: Teaching a Machine Its Own If-Else
  2. Counting the Knobs: Parameters, Bits, Training and Inference
  3. A Quick Peek Inside the Neuron
  4. Which One Is It? Classification and Softmax
  5. Networks Everywhere: How Machines Learned to See
  6. Feeding Words to Numbers: Tokens and Bag of Words
  7. Words in Order: Recurrent Networks and LSTM
  8. The Big Idea: Predict the Next Word