Add layers, tune neurons, change activation functions — then watch forward propagation light up your network with real computed values at every node!
Each neuron receives inputs, multiplies by weights, adds a bias, then passes through an activation function.
Every neuron in one layer connects to every neuron in the next layer. Each connection has a learnable weight.
Data flows left to right through the network, transforming at each layer, until it reaches the output.
Error flows backwards from output to input, adjusting weights to reduce mistakes. Repeat thousands of times!
You built a neural network and watched it learn through backpropagation!
Each connection has a weight; each neuron has a bias. Training adjusts all of these to minimise the error.
ReLU, Sigmoid, Tanh — without these, a 1000-layer network is mathematically the same as 1 linear layer.
Data multiplied by weights, summed, activated — repeated layer by layer until the output neuron.
Mean Squared Error (MSE) quantifies how wrong the output is. Training minimises this number.
Using the chain rule of calculus to compute how much each weight contributed to the error, then adjusting it.
Take small steps in the direction of steepest descent on the loss surface. Learning rate controls the step size.