2.3 KiB
2.3 KiB
| title | date | tags | techne | episteme | slug | |
|---|---|---|---|---|---|---|
| [SI] Some Questions | 2012-01-11 |
|
:done | :speculation | 2012/01/11/si-some-questions/ |
Let's kick off the thinking process about Solomonoff Induction (SI), Kolmogorov complexity (KC), AIXI and so on. Here's some questions I wanna answer.
- How do you calculate the KC of a program?
- Why are KC and SI incomputable? How bad is this limitation?
- Are there alternatives to KC? Are there different variants?
- How do you actually predict something with SI?
- How does SI work with non-sequential data? (i.e., normally SI completes a sequence of bits, but we also wanna use it for tasks like "Is this sentence part of this language?")
- How does AIXI work?
- What are computable approximations of KC, SI and AIXI?
- Are there existing practical uses for SI? If there are alternative approaches for a use case, how well does SI fare? (Is it as awesome as Bayes?)
- What's the relationship to Gödel's Incompleteness Theorem (GIT)?
- Can a human use some form of SI? If not, why not? If so, examples? (Compare with Bayes, which is a bit tricky too use for humans with larger data sets, but is still usable, as Carrier's resurrection arguments show.)
- Do SI and AIXI mostly rely on brute computation or memory or both? (Or is it unknown?)
- SI is a complexity-only prior for Occam's Razor. Are there speed priors? Do they have some obvious disadvantages? (Has SI?)
- KC can be used to test a string for its randomness. How does that work?
- Using KC, is there a difference between real randomness and pseudo-randomness?
- What's some recent stuff that's happening? Is the research making progress? Does anyone care about SI besides some math heads?
- Are there some obvious philosophical implications of SI?
- (What's up with Ray Solomonoff's beard? I mean, seriously.)
This is gonna be an interesting week. As I said, this is my thinking process - I can't answer many of these questions myself yet! Once I have written this stuff and have a clear picture, I'll clean it up and turn it into an article (or short sequence, if it's too long). Then feedback, improvements, karma. Or I end up hating math forever. It's an adventure! (I'm getting my wisdom teeth removed tomorrow, so if this feels particularly incoherent, I blame the meds.)