Re:Ranked programming: Probabilistic programming without pro...
by SiempreViernes on Friday November 20, 2020 @04:54AM
(#25158936)
While I didn't read that article in detail, I did not find any clear discussion of how it differs from probability.
The one difference I saw was that total possibility is not normalised, in principle for a continuous variable the total can be infinity. But I'm not sure what effect this has in practice besides making it tricky to work with some contrived examples.
Re:Ranked programming: Probabilistic programming without pro...
by anentropic on Friday November 20, 2020 @06:26AM
(#25159408)
I really have no idea... but what I think I understood is:
Probabilities have a magnitude, i.e. 0.5 is twice as likely as 0.25
But "surprise ranks" only have an ordering: 0 (inevitable) < 1 < 2 < 3 ... < inf (impossible)
It can be useful when the actual probabilities of events are not known (it's often the case in real life scenarios). Then possibility theory may actually tell you what decision to make based on this limited information. My understanding is that the point is not that probability theory does not work, but that it requires a lot of knowledge about the system, which might be not available to the agent. In such situations, possibility theory provides a framework for making reasonable deductions.