Ok ok, so I haven’t really built a Predictatron in a box. But I really have developed and implemented a very general learning algorithm, called Blerpl. More about that soon.
The point of the switches and lights story was to frame the prediction problem in a very specific way. I want you to imagine stepping through the process of reading out predictions and feeding in observations one at a time. I want the data format to be binary, and for the interface to be completely generic, with no preconceived notions about the world that the learning algorithm is exposed to. And I want you to believe that, reduced to such a simple description, this might be a problem we can solve with computers in a very general way. There need be no magic involved.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment