Understanding Neural Networks: From Neurons to Multi-Layer Perceptrons 🧠

"What is a neuron and how does it work? From a single neuron to a layer of neurons to multiple layers of neurons."

___________________________________________
Subscribe the channel https://www.youtube.com/channel/UCgQlZ6kefvYeHDe__YkFluA/?sub_confirmation=1

___________________________________________
Part 1: Why Neural Networks for Machine Learning?
https://www.youtube.com/watch?v=NaYvohpr9No

Part 2: Building Neural Networks - Neuron, Single Layer Perceptron, Multi Layer Perceptron
https://www.youtube.com/watch?v=MqeGZcqkrn0

Part 3: Activation Function of Neural Networks - Step, Sigmoid, Tanh, ReLU, LeakyReLU, Softmax [
https://www.youtube.com/watch?v=rtinxohdo7Y

Part 4: How Neural Networks Really Work - From Logistic to Piecewise Linear Regression
https://www.youtube.com/watch?v=rtinxohdo7Y

Part 5: Delta Rule for Neural Network Training as Basis for Backpropagation
https://www.youtube.com/watch?v=wKbZkfuuQLw

Part 6: Derive Backpropagation Algorithm for Neural Network Training
https://www.youtube.com/watch?v=s1CFVeJHmQk

Part 7: Gradient Based Training of Neural Networks
https://www.youtube.com/watch?v=N8ZKqB19Vw4

___________________________________________
In this section, we build neural networks starting from a signle neuron. The input variables are scaled with multiplicative weights. An added bias term leads to the linear in a neuron. This result is modified non-linearly by an activation function. For a sigmoid activation function, a signle neuron is equivalent to logistic regression.
Placing neurons in parallel lead to a single layer perceptron (SLP). Stacking such layers of neurons together gives a multiple layer perceptron (MLP). The layers apart from the output layer are called hidden layers.The calculations in such neural networks can be written as matrix-vector multiplications. In this lecture, we consider fully connected feedforward neural networks without reconnecting neurons or shortcut connections.