Physics simulations for virtual smoke, explosions or water are by now crucial tools for special effects. Despite their widespread use, it is still difficult to get get these simulations under control, and they are still far too expensive for practical interactive applications. In this talk I will outline research directions to alleviate these inherent difficulties with machine learning techniques based on neural networks. As powerful tools to approximate complex nonlinear functions these networks can enable new directions of working with fluid simulations. I will show several recent examples of how fluid simulations and neural networks can work together, e.g., to synthesize new simulations with pre-computed patches of flow data, or to enable interactive liquids applications by warping space-time surfaces. In the end, I will also discuss possible future directions for this area. I think fluids (as complex physics phenomena) pose very interesting research challenges for neural networks.