Artificial Neural Networks

The Perceptron

• Simplest ANN
• Based upon - Linear Threashold Unit (LTU). In neurons, you give signals. In perceptron you give numbers.
• Each connection is associated with weight
• LTU computes weighted sum of inputs

• A single LTU can be used for simple linear binary classification
• Computes linear combination of inputs, & if result is more than a threshold - it gives positive otherwise negative
• A perceptron is single layer of LTU, with each neuron connected to all inputs.

• Perceptron do not output class probalility, they just make hard prediction.
• This limitation of perceptron can be entirely eleminated by stacking multiple perceptrons

Multi-Layer Perceptron & Backpropagation

Deep Neural Networks

• Having multiple hidden layers