In an already crowded space, what does this bring me that other libraries don't?

(Reading between the lines, is it something to do with avoiding NVIDIA CUDA? In which case performance comparisons would be really helpful)

In context, ML makes great sense as an acronym. The problem is a headline on a news aggregator hasn't much context.

For the curious, this is about a Machine Learning library, not a library for Standard ML, ML with Concurrency, CAML, OCAML, or F#.