In the GDC talk it was prominently mentioned that Houdini Vellum Simulations where used to create training Data for the ML Deformer.
I cannot find any ressources, tutorials, documention that explains the process.
Is there going to be a Houdini ML export PlugIn in the future like there is for maya or Documentation on how to achieve this?
The Houdini setup will ship as part of the next Houdini release from what I understand.
So it is coming 
Our ultimate goal is to do all simulation and training in engine so you won’t have to hop between different DCC’s anymore. The Chaos ML Cloth in engine training data generation is one step towards that.
There are currently no plans for any plugin for Houdini. The plugin is only to generate random poses though. I hope we can also bring that pose generation also inside the engine later on.