Ordinary differential equations (ODEs) are of paramount importance in the modeling of biochemical and cyber-physical systems. Unfortunately, many practical systems are too large to be analyzed efficiently. A common way of approaching this problem are reduction (alternatively, abstraction) techniques which relate the original system to a smaller one in a formal sense. However, at the same time, it is well-known that common notions of reduction are not robust to parameter perturbation. After discussing exact notions of reduction, the present talk will therefore address approximate reduction techniques. Moreover, future research directions will be addressed.