I would definitely call it an open research problem to provide PAC-style bounds for more complicated hypothesis spaces and learning settings. But that doesn’t mean it’s impossible or un-doable, just that it’s an open research problem. I want a limitary theorem proved before I go calling things impossible.
I would definitely call it an open research problem to provide PAC-style bounds for more complicated hypothesis spaces and learning settings. But that doesn’t mean it’s impossible or un-doable, just that it’s an open research problem. I want a limitary theorem proved before I go calling things impossible.