Seite - 451 - in Differential Geometrical Theory of Statistics
Bild der Seite - 451 -
Text der Seite - 451 -
Entropy2016,18, 110
3.4. KolmogorovComplexityand theAsymptoticBound
Werecallherebrieflytheresultof [16] linkingtheasymptoticboundoferrorcorrectingcodes to
Kolmogorovcomplexity.
Aswediscussedabove, only theasymptoticboundmarksa significant changeofbehaviorof
codesacross thecurve (isolatedcodepointswithfinitemultiplicityversusaccumulationpointswith
infinitemultiplicity). In thissense thiscurve isverydifferent fromall theotherbounds in thespace
of codeparameters.However, there isnoexplicit expression for thecurveR= αq(δ) thatgives the
asymptoticbound. Indeed,eventhequestionof thecomputabilityof the functionR=αq(δ) isapriori
unclear. Thisquestionwasformulatedexplicitly in [25].
It isproved in [16] that theasymptoticboundR= αq(δ)becomescomputablegivenanoracle
thatcan list codesbyincreasingKolmogorovcomplexity.Givensuchanoracle,onecanprovidean
explicit iterative (algorithmic)procedure forconstructing theasymptoticbound. This implies that the
asymptoticboundis“atworstasnon-computableasKolmogorovcomplexity”.
Consider thesetX=Cqof (unstructured)q-arycodesandthesetY⊂ [0,1]2 ofcodepointsand
thecomputable function f :X→Y thatassigns toacodeC∈X its codeparameters (R(C),δ(C))∈Y.
LetYfin andY∞be, respectively, thesubsetsof thespaceofcodepoints thatcorrespondtocodepoints
realizedwithfinite andwith infinitemultiplicity. Thealgorithm iterativelyproduces twosets Am
andBm thatapproximate, respectively,Y∞ andYfinbyYfin=∪m≥1Bm andY∞=∪m≥1(∩n≥0Am+n).
The inductive construction starts by choosing an increasing sequenceof positive integersNm and
settingB1=∅andtakingA1 tobe thesetofcodepointsywithν−1Y (y)≤N1,whereνY :N→Y isa
fixedenumerationof thesetof rationalpoints [0,1]2∩Q2wherecodepointsbelong.
General estimates on the behavior of (exponential) Kolmogorov complexity under
composition of total recursive functions (see [30], Section VI.9 of [32]) show that, for
a composition F= f0(f1(t1, . . . ,tm), · · · , fn(t1, . . . ,tm),tm+1, . . . ,t ) of recursive functions the
Kolmogorovcomplexitysatisfies
K(F)≤ c · n
∏
i=1 K(fi) · (
log n
∏
i=1 K(fi) )n−1
,
forafixed f0 andvarying fi, i≥1.
Consider the total recursive functionF(x)=(f(x),n(x))with
n(x)=#{x′ |ν−1X (x′)≤ ν−1X (x), f(x′)= f(x)}
where νX :N→X is an enumerationof the spaceof codes. Consider the enumerable setsXm :=
{x ∈ X |n(x) = m} and Ym := f(Xm) ⊂ Y, with Y∞ = ∩mf(Xm) and Yfin = f(X) Y∞. For
ϕ : f(X)→ X1, defined as f−1 on f(X1) = f(X), applying the composition rule for exponential
Kolmogorovcomplexity, it is showninProposition3.1of [16] that, forx∈X1 andy= f(x), onehas
K(x)=K(ϕ(y))≤ cϕ ·K(y)≤ cν−1Y (y), hence
KTU(x)≤ c ·ν−1Y (y).
UsingthesametypeofestimateofKolmogorovcomplexityforcompositionofrecursivefunctions,
it is thenshown inProposition3.2 [16] that, fory∈Y∞ andm≥ 1, and foraunique xm ∈X,with
y= f(xm),n(xm)=mand c= c(f,u,v,νX,νY)>0,onefinds
KTU(xm)≤ c ·ν−1Y (y)m log(ν−1Y (y)m).
451
Differential Geometrical Theory of Statistics
- Titel
- Differential Geometrical Theory of Statistics
- Autoren
- Frédéric Barbaresco
- Frank Nielsen
- Herausgeber
- MDPI
- Ort
- Basel
- Datum
- 2017
- Sprache
- englisch
- Lizenz
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03842-425-3
- Abmessungen
- 17.0 x 24.4 cm
- Seiten
- 476
- Schlagwörter
- Entropy, Coding Theory, Maximum entropy, Information geometry, Computational Information Geometry, Hessian Geometry, Divergence Geometry, Information topology, Cohomology, Shape Space, Statistical physics, Thermodynamics
- Kategorien
- Naturwissenschaften Physik