First of all, this sounds weird—Solomonoff induction isn’t about scoring two hypothesis. It includes something which does (something like) that, and then it does that for all possible programs/hypothesis (which is one reason why people are saying it’s uncomputable*) and then it has a universal prior, and as it gets new evidence it updates that (infinite) prior probability distribution (which is the second reason why it’s uncomputable*).
AIXI is related. Being based on Solomonoff induction, it is also incomputable. However, there’s AIXItl which is an approximation (with memory and time contraints), and it’s okay at something like first person pac-man, so there probably is an approximation to Solomonoff induction. I don’t know how useful it is, and I’ve never seen it used.
Yes, this is helpful—I had thought of Solomonoff induction as only being calculating the prior but it’s helpful to understand the terminology properly.
First of all, this sounds weird—Solomonoff induction isn’t about scoring two hypothesis. It includes something which does (something like) that, and then it does that for all possible programs/hypothesis (which is one reason why people are saying it’s uncomputable*) and then it has a universal prior, and as it gets new evidence it updates that (infinite) prior probability distribution (which is the second reason why it’s uncomputable*).
*a.k.a.: this takes literally forever.
AIXI is related. Being based on Solomonoff induction, it is also incomputable. However, there’s AIXItl which is an approximation (with memory and time contraints), and it’s okay at something like first person pac-man, so there probably is an approximation to Solomonoff induction. I don’t know how useful it is, and I’ve never seen it used.
Sure. The OP might more accurately have asked “How is the Solomonoff prior calculated?”
Yes, this is helpful—I had thought of Solomonoff induction as only being calculating the prior but it’s helpful to understand the terminology properly.