I toyed around with this 7 years ago after having first done an extensive research survey. It's pretty straightforward in theory and there are numerous approaches that can be combined. At that time there was no research article showing the full combination of techniques. The capabilities of the various forwarding chips potentially limits the full scope of what's possible. My early simulations were quite promising. The effort has been enjoying project shelf/vacation since then.
One of the main downsides with the approach is that it is very hard to get a feel for how well the forwarding table can compress, and you are more than usually at the mercy of what your peers are sending you.
Best,
Martin