It's really difficult to work with tabular/structured data on these Deep Nets. Particularly when the structured data has a variable number of entries. This has

It's really difficult to work with tabular/structured data on these Deep Nets. Particularly when the structured data has a variable number of entries. This has less to do with theoretical constraints, and more to do with the mechanics of our current ML toolchains. I'd imagine if the toolchain problem was solved we'd see a lot more research in this direction.