Yet another attempt to own everything. They even reference Martin Riedmiller's work from 2005 on "Neural Fitted Q Iteration" and dismiss it, saying that his pa

Yet another attempt to own everything. They even reference Martin Riedmiller's work from 2005 on "Neural Fitted Q Iteration" and dismiss it, saying that his paper was just an "attempt" to use an MLP for the Q function, and that because his paper talks about rebuilding a new network with each new batch of data, resulting in an inefficient learning process - ergo, their method is novel. What utter nonsense. This completely misrepresents Riedmiller's work. They are really splitting some fine hairs here. Also, they admit that Martin Riedmiller wrote about the idea, even though he didn't use it in that paper. And yet, they still hope to control the technology.