The ability to have all inputs available to all state machines is really useful - they also mention a port of tensorflow lite to the board. So, one example would be preprocessing inputs for feeding directly into tensorflow applications, for ML at the edge (I think that's the buzzword bingo won there!).

With 360mb/s I/O bandwidth, that equates to more than FullHD at 60hz, with 16bit colour depth. If you attach say a rpi camera to the I/O you can do signal processing on the image. Same for VGA output.