请输入您要查询的字词:

 

单词 4Measurement
释义

4. Measurement


This sectionPlanetmathPlanetmathPlanetmathPlanetmathPlanetmath adapts Definition 1 (http://planetmath.org/1introduction#Thmdefn1)to distributedstochastic systems. The first step is to replace elements ofstate spacePlanetmathPlanetmath X with stochastic maps din:𝒱S𝐃, or equivalently probability distributionson S𝐃, which are the system’s inputs. Individualelements of S𝐃 correspond to Dirac distributions.

Second, replace function f:X with mechanism𝔪𝐃:𝒱S𝐃𝒱A𝐃.Since we are interested in the compositional structureMathworldPlanetmath ofmeasurements we also consider submechanisms 𝔪𝐂.However, comparing mechanisms requires that they have the samedomain and range, so we extend 𝔪𝐂 to the entiresystem as follows

𝔪𝐂=𝒱S𝐃𝜋𝒱S𝐂𝔪𝐂𝒱A𝐂π𝒱A𝐃.(1)

We refer to the extensionPlanetmathPlanetmath as 𝔪𝐂 by abuse of notation.We extend mechanisms implicitly whenever necessary withoutfurther comment. Extending mechanisms in this way maps thequale into a cloud of points in Hom(𝒱A𝐃,𝒱S𝐃) labeled by objects in 𝚂𝚢𝚜𝐃.

In the special case of the initial objectMathworldPlanetmath 𝐃, define

𝔪=𝒱S𝐃𝜔ω𝒱A𝐃.
Remark 3.

Subsystems differing by non-existent edges (Remark 2 (http://planetmath.org/3distributeddynamicalsystems#Thmrem2))are mapped to the same mechanism by this construction, thusmaking the fact that the edges do not exist explicit withinthe formalism.

Composing an input with a submechanism yields an outputdout:=𝔪𝐂din:𝒱A𝐃, which is a probability distribution onA𝐃. We are now in a position to define

Definition 8.

A measuring device is the dual 𝔪𝐂to the mechanism of a subsystem. An output is astochastic map dout:𝒱A𝐃. Ameasurement is a composition 𝔪𝐂dout:𝒱S𝐃.

Recall that stochastic maps of the form 𝒱X correspond to probability distributions on X. Outputs asdefined above are thus probability distributions onA𝐃, the output alphabetMathworldPlanetmath of 𝐃. Individualelements of A𝐃 are recovered as Dirac vectors:δa𝒱A𝐃.

Definition 9.

The effective information generated by 𝐂1in the context of subsystem 𝐂2𝐂1 is

ei(𝔪𝐂2𝔪𝐂1,dout):=H[𝔪𝐂1dout𝔪𝐂2dout].(2)

The null context, corresponding to the empty subsystem=V𝐃×V𝐃, is a specialcase where 𝔪𝐂dout is replaced by theuniform distributionMathworldPlanetmath ω𝐃 on S𝐃.To simplify notation define

ei(𝔪𝐂,dout):=ei(𝔪𝔪𝐂,dout).

Here, H[pq]=ipilog2piqi is theKullback-Leibler divergence or relative entropyMathworldPlanetmath[1]. Eq. (2) expands as

ei(𝔪𝐂2𝔪𝐂1,dout)=sS𝐃𝔪𝐂1dout|δslog2𝔪𝐂1dout|δs𝔪𝐂2dout|δs.(3)

When dout=δa for some aA𝐃 wehave

ei(𝔪𝐂2𝔪𝐂1,δa)=sS𝐃p𝔪𝐂1(s|a)log2p𝔪𝐂1(s|a)p𝔪𝐂2(s|a).(4)

Definition 8 requires some unpacking. To relateit to the classical notion of measurement, Definition 1 (http://planetmath.org/1introduction#Thmdefn1),we consider system 𝐃={vX𝑓vY}where the alphabets of vX and vY are the setsAvX=X and AvY=Y respectively, and themechanism of vY is 𝔪Y=𝒱f. In other words,system 𝐃 corresponds to a single deterministicMathworldPlanetmath functionf:XY.

Proposition 5 (classical measurement).

The measurement (Vf)δyperformed when deterministic function f:XYoutputs y is equivalentMathworldPlanetmathPlanetmathPlanetmathPlanetmathPlanetmath to the preimageMathworldPlanetmath f-1(y).Effective information is ei(Vf,δy)=log2|X||f-1(y)|.

Proof:By Corollary 2 (http://planetmath.org/2stochasticmaps#Thmthm2)measurement (𝒱f)δy is conditionaldistribution

p𝒱f(x|y)={1|f-1(y)|if f(x)=y0else.

which generalizes the preimage. Effective information followsimmediately.

Effective information can be interpreted as quantifying ameasurement’s precision. It is high if few inputs cause fto output y out of many – i.e. f-1(y) has fewelements relative to |X| – and conversely is low if manyinputs cause f to output y – i.e. if the output isrelatively insensitive to changes in the input. Precisemeasurements say a lot about what the input could have beenand conversely for vague measurements with low ei.

The point of this paper is to develop techniques for studyingmeasurements constructed out of two or more functions. Wetherefore present computations for the simplest case,distributed system X×Y𝑔Z, in considerabledetail. Let 𝐃 be the graph

with obvious assignments of alphabets and the mechanism ofvZ as 𝔪Z=𝒱g. To make the formulasMathworldPlanetmathPlanetmath more readablelet 𝔪XY=𝒱g, 𝔪X=𝒱gπXY,X and 𝔪Y=𝒱gπXY,Y. We then obtain latticeMathworldPlanetmath

The remainder of this section and most of the next analyzesmeasurements in the lattice.

Proposition 6 (partial measurement).

The measurement performed on X when g:X×YZ outputs z, treating Y as extrinsicnoise, is conditional distribution

p(x|z)={|gx×Y-1(z)||g-1(z)|if g(x,y)=z for some yY0else,(5)

where gx×Y-1(z):=prY(g-1(z){x}×Y). The effective information generated by thepartial measurement is

ei(𝔪X,δz)=log2|X|+xXp(x|z)log2p(x|z).δz)|g-1(z)|(6)

Proof: Treating Y as a source of extrinsic noise yields𝒱Xπ𝒱X𝒱Y𝒱g𝒱Z which takesδx1|Y|yYδg(x,y).The dual is

𝔪X=πXY,X(𝒱g):δzxX|gx×Y-1(z)||g-1(z)|δx.

The computation of effective information follows immediately.

A partial measurement is precise if the preimage g-1(z)has small or empty intersectionMathworldPlanetmathPlanetmath with {x}×Y for mostx, and large intersection for few x.

PropositionsPlanetmathPlanetmath 5 and 6 computeeffective information of a measurement relative to the nullcontext provided by completePlanetmathPlanetmathPlanetmathPlanetmathPlanetmath ignorance (the uniformdistribution). We can also compute the effective informationgenerated by a measurement in the context of a submeasurement:

Proposition 7 (relative measurement).

The information generated by measurement X×Y𝑔Z in the context of the partial measurement where Y is unobserved noise, is

ei(𝔪X𝔪XY,δz)=xXgx×Y-1(z)g-1(z)log2|Y|gx×Y-1(z).(7)

Proof: Applying Propositions 5 and6 obtains

ei(𝔪X𝔪XY,δz)=(x,y)g-1(z)1|g-1(z)|log2[1|g-1(z)||g-1(z)||Y||gx×Y-1(z)|]

which simplifies to the desired expression.

To interpret the result decompose X×Y𝑔Zinto a family of functions ={Ygx×YZ|xX}labeled by elements of X, where gx×Y(y):=g(x,y).The precision of the measurement performed by gx×Ys log2|Y|gx×Y-1(z). It follows thatthe precision of the relative measurement,Eq. (7), is the expected precision of themeasurements performed by family taken withrespect to the probability distributionp(x|z)=gx×Y-1(z)g-1(z) generated bythe noisy measurement.

In the special case of g:X×YZ relativeprecision is simply the difference of the precision of thelarger and smaller subsystems:

Corollary 8 (comparing measurements).
ei(𝔪X𝔪XY,δz)=ei(𝔪XY,δz)-ei(𝔪X,δz)

Proof: Applying Propositions 5,6, 7 and simplifying obtains

ei(𝔪XY,δz)-ei(𝔪X,δz)=log2|X||Y||g-1(z)|-x|gx×Y-1(z)||g-1(z)|log2|X||gx×Y-1(z)||g-1(z)|
=log2|Y||g-1(z)|+(x,y)g-1(z)1|g-1(z)|log2|g-1(z)||gx×Y-1(z)|
=ei(𝔪X𝔪XY,δz).

References

  • 1 E T Jaynes (1985):EntropyMathworldPlanetmath and Search Theory. In CR Smith & WT Grandy,editors: Maximum-entropy and Bayesian Methods inInverse Problems, Springer.
随便看

 

数学辞典收录了18232条数学词条,基本涵盖了常用数学知识及数学英语单词词组的翻译及用法,是数学学习的有利工具。

 

Copyright © 2000-2023 Newdu.com.com All Rights Reserved
更新时间:2025/5/4 20:30:30