The function would simply be a chain of multiplications and (typically) non-linear transformations.
For classification problems, there is often a final non-linear transformation, like a softmax.
Example with ReLu activation:
Output_i = max((input_i * weight_i),0)