Sounds interesting, nice to see that there is research in that direction. Though it will be even harder for some new format to replace IEEE floats than with e.g. new image formats like APNG or FLIF. The old IEEE floats have many flaws/pitfalls, but they have been used and researched for a long time. After a quick glance, it looks like a useful experimental new format that seems to focus on precision instead of performance (although it tries to adress this, too).
http://schnaader.info
Damn kids. They're all alike.
I think its mainly targeted at NN/TPU (8-bit floats etc).
I've been keeping track of Unums – it's very interesting. I'm also a fan of logarithmic number systems. Like this one.