Feed Forward Neural: Learning Parameters Intuition

How can we go about learning the parameters of a feedforward neural network? Remember that the gradient descent algorithm for a simple neural network was as follows:

Now, instead of $W$ and $b$, we have $[W_1,W_2,...,W_L]$ and $[b_1,b_2,...,b_L]$. We can put both of these into one vector called $\theta$, modifying the algorithm to:

Where:

$$\theta =[W_1,W_2,...,W_L,b_1,b_2,...,b_L]$$ $$\nabla {\theta }_t=[\frac{\partial \mathcal{L}(\theta )}{\partial W_{1,t}},...,\frac{\partial \mathcal{L}(\theta )}{\partial W_{L,t}},\frac{\partial \mathcal{L}(\theta )}{\partial b_{1,t}},...,\frac{\partial \mathcal{L}(\theta )}{\partial b_{L,t}}]$$

$\nabla \theta$ is composed of the gradients of the weight and bias of each layer in the network. So, how can we calculate the loss function and how can we calculate the gradient?