The gradient is the vector of partial derivatives of the loss function with respect to each parameter, pointing in the direction of steepest increase — training moves parameters in the opposite direction. Gradient-based optimization underpins virtually all modern deep learning: backpropagation computes gradients efficiently through the chain rule, and optimizers like SGD or Adam use those gradients to update weights. Issues like vanishing or exploding gradients can stall or destabilize training, motivating architectural choices such as residual connections and layer normalization.