Solomonoff induction
E774592
formal theory of universal prediction
idealized prediction method
incomputable prediction scheme
theory in algorithmic information theory
Solomonoff induction is a formal theory of universal prediction that combines algorithmic information theory and Bayesian reasoning to define an idealized, incomputable method for inferring future data from past observations.
Statements (49)
| Predicate | Object |
|---|---|
| instanceOf |
formal theory of universal prediction
ⓘ
idealized prediction method ⓘ incomputable prediction scheme ⓘ theory in algorithmic information theory ⓘ |
| aimsTo | predict future observations from past data ⓘ |
| appliesTo |
online learning scenarios
ⓘ
sequential prediction problems ⓘ |
| approximatedBy |
MDL-based predictors
ⓘ
compression-based prediction methods ⓘ resource-bounded variants ⓘ |
| assumes | data generated by a computable process ⓘ |
| basedOn |
Bayesian mixture of all programs
ⓘ
mixture over all computable hypotheses ⓘ |
| componentOf | AIXI formalism NERFINISHED ⓘ |
| defines | universal a priori probability distribution ⓘ |
| encodes | Occam’s razor via program length prior ⓘ |
| field |
Bayesian statistics
ⓘ
algorithmic information theory ⓘ machine learning theory ⓘ philosophy of science ⓘ |
| formalizedAs | universal semimeasure M NERFINISHED ⓘ |
| goal | provide a gold standard for inductive inference ⓘ |
| guarantees | convergence to true computable distribution ⓘ |
| influenced |
theory of inductive inference
ⓘ
universal artificial intelligence ⓘ |
| inspired | AIXI NERFINISHED ⓘ |
| introducedBy | Ray Solomonoff NERFINISHED ⓘ |
| limitation | not computable in practice ⓘ |
| minimizes |
expected log-loss asymptotically
ⓘ
expected number of prediction errors asymptotically ⓘ |
| namedAfter | Ray Solomonoff NERFINISHED ⓘ |
| property |
computably enumerable from below
ⓘ
computationally incomputable ⓘ |
| publication | A Formal Theory of Inductive Inference NERFINISHED ⓘ |
| publicationYear | 1964 ⓘ |
| relatedTo |
Bayesian universal coding
NERFINISHED
ⓘ
Kolmogorov complexity-based learning ⓘ Minimum Description Length principle NERFINISHED ⓘ |
| satisfies |
asymptotic optimality in sequence prediction
ⓘ
universal dominance over all computable semimeasures ⓘ |
| uses | prefix-free Turing machine programs ⓘ |
| usesConcept |
Bayesian updating
ⓘ
Epicurus’ principle of multiple explanations NERFINISHED ⓘ Kolmogorov complexity NERFINISHED ⓘ Occam’s razor NERFINISHED ⓘ algorithmic probability ⓘ universal Turing machine NERFINISHED ⓘ universal prior ⓘ |
| weights | shorter programs more heavily ⓘ |
Referenced by (5)
Full triples — surface form annotated when it differs from this entity's canonical label.
Universal Intelligence: A Definition of Machine Intelligence
→
influencedBy
→
Solomonoff induction
ⓘ