Machine Learning in Python

Neurons receive preprocessed features as inputs, similar to how 

biological receptors respond to stimuli like light. Each neuron processes these inputs to 

produce a singular output.

Weights in neurons adjust during learning, similar to tuning 

synapses in the brain. These adjustments fine-tune the neuron's output to match the 

desired outcome of the network's training. 

The bias is a crucial parameter that adds flexibility to a neuron's output, 

allowing it to activate effectively even with no input. It helps the network fit the data 

better, enabling the modelling of complex functions and decision boundaries. 

This function calculates the weighted sum of inputs and 

weights, then adds the bias, setting the stage for the neuron's activation. 

Transforms the summation output into a complex, non-linear 

form, allowing for neuron activation or deactivation. It's essential for enabling multi-layer 

networks to learn beyond linear classification, with common types including Sigmoid, 

Tanh, ReLU, and Softmax. 

 The result from a neuron's internal processing, including weighted inputs 

and bias through an activation function, becomes the output, which then serves as input 

to subsequent layers in the network. 

The initial layer of a neural network 

directly interfaces with the input data, 

consisting of multiple nodes that 

correspond to the features of the input. 

For example, a 28x28 pixel image could 

have an input layer with 784 nodes. 

Each node represents a single feature or 

pixel value, transmitting this information 

unchanged to the hidden layers of the 

network for further processing.