## Error Estimations For Indirect Measurements: Randomized Vs. Deterministic Algorithms For "Black-Box" Programs (2000)

Venue: | Handbook on Randomized Computing, Kluwer, 2001 |

Citations: | 29 - 13 self |

### BibTeX

@INPROCEEDINGS{Kreinovich00errorestimations,

author = {Vladik Kreinovich and Raúl Trejo and Ra Ul Trejo},

title = {Error Estimations For Indirect Measurements: Randomized Vs. Deterministic Algorithms For "Black-Box" Programs},

booktitle = {Handbook on Randomized Computing, Kluwer, 2001},

year = {2000},

pages = {673--729},

publisher = {Kluwer}

}

### Years of Citing Articles

### OpenURL

### Abstract

In many real-life situations, it is very difficult or even impossible to directly measure the quantity y in which we are interested: e.g., we cannot directly measure a distance to a distant galaxy or the amount of oil in a given well. Since we cannot measure such quantities directly, we can measure them indirectly: by first measuring some relating quantities x1 ; : : : ; xn , and then by using the known relation between x i and y to reconstruct the value of the desired quantity y. In practice, it is often very important to estimate the error of the resulting indirect measurement. In this paper, we describe and compare different deterministic and randomized algorithms for solving this problem in the situation when a program for transforming the estimates e x1 ; : : : ; e xn for x i into an estimate for y is only available as a black box (with no source code at hand). We consider this problem in two settings: statistical, when measurements errors \Deltax i = e x i \Gamma x i are inde...

### Citations

823 |
Fuzzy Sets and Fuzzy Logic – Theory and Applications
- Klir, Yuan
- 1995
(Show Context)
Citation Context ...iate degrees of belief). This idea was first proposed by L. Zadeh [66] under the name of fuzzy set theory. There are many different procedures that enable to obtain these numerical values (see, e.g., =-=[19, 32]-=-). If we apply any of these procedures, then for every e, we get the degree of belief (e) that this e satisfies the above statement. In mathematical terms, we get a function from the set of all real n... |

513 |
Fuzzy Sets and Systems: Theory and Applications
- Dubois, Prade
- 1980
(Show Context)
Citation Context ...iate degrees of belief). This idea was first proposed by L. Zadeh [66] under the name of fuzzy set theory. There are many different procedures that enable to obtain these numerical values (see, e.g., =-=[19, 32]-=-). If we apply any of these procedures, then for every e, we get the degree of belief (e) that this e satisfies the above statement. In mathematical terms, we get a function from the set of all real n... |

356 |
Lectures on Functional Equations and their Applications
- Aczél
- 1966
(Show Context)
Citation Context ... 2 , so we can conclude that (c 1 + c 2 ) = (c 1 ) \Delta (c 2 ). It is known that the only measurable function which satisfies this functional equation is (!) = exp(b \Delta !) for some real value b =-=[1]-=-. (The proof is that 's logarithm is an 20 additive function, and it has been, basically, known since Cauchy that the only measurable additive functions are linear ones.) Hence, for ! ? 0, we have (!)... |

351 |
Possibility Theory: An Approach to Computerized Processing of Uncertainty
- Dubois, Prade
- 1988
(Show Context)
Citation Context ... who introduced them in [17, 18]. They also described how to apply arithmetic operations to these "fuzzy numbers". For detailed expositions of LR-numbers, see [19], Section II.2.B; [30], Sec=-=tion 1.9; [20]-=-, Chapter 2 and Section 3.2.4; [69], Chapter 5. Let's describe their main formulas in application to our problem. How to Describe a Resulting Expert Estimate For \Deltay? Extension Principle For LR-Nu... |

329 |
Measurement Error Models
- Fuller
- 1987
(Show Context)
Citation Context ... only know intervals will be one of the settings considered in this paper. 4 Another setting which we will consider is a setting described in standard engineering textbooks on measurement (see, e.g., =-=[25, 56]-=-; see also [13, 28]). In this setting, the measurement error \Deltax i of each direct measurement is normally distributed with 0 average and known standard deviation oe i , and measurement errors of d... |

295 |
Evaluating derivatives: principles and techniques of algorithmic differentiation
- Griewank
- 2000
(Show Context)
Citation Context ...nalytical expression. When this algorithm is presented in one of the standard programming languages such as Fortran or C, we can apply one of the existing analytical differentiation tools (see, e.g., =-=[8, 27]-=-), and automatically produce a program which computes the partial derivatives c i . These tools analyze the code and produce the differentiation code as they go. In Many Practical Applications, We Mus... |

185 | Computational Complexity and Feasibility of Data Processing and Interval Computations
- Kreinovich, Lakeyev, et al.
- 1997
(Show Context)
Citation Context ... and references therein). In this case, the problem of error estimation for indirect measurements becomes computationally difficult (NP-hard) even when the function f(x 1 ; : : : ; x n ) is quadratic =-=[43, 62]-=-. However, in most real-life situations, the possibility to ignore quadratic terms is a reasonable assumption, because, e.g., for an error of 1% its square is a negligible 0.01%. With the above restri... |

125 |
Nonlinear optimization. Complexity issues
- Vavasis
- 1991
(Show Context)
Citation Context ... and references therein). In this case, the problem of error estimation for indirect measurements becomes computationally difficult (NP-hard) even when the function f(x 1 ; : : : ; x n ) is quadratic =-=[43, 62]-=-. However, in most real-life situations, the possibility to ignore quadratic terms is a reasonable assumption, because, e.g., for an error of 1% its square is a negligible 0.01%. With the above restri... |

124 |
Introduction to fuzzy arithmetic: Theory and application
- Kaufmann, Gupta
- 1991
(Show Context)
Citation Context ...Dubois and H. Prade who introduced them in [17, 18]. They also described how to apply arithmetic operations to these "fuzzy numbers". For detailed expositions of LR-numbers, see [19], Sectio=-=n II.2.B; [30]-=-, Section 1.9; [20], Chapter 2 and Section 3.2.4; [69], Chapter 5. Let's describe their main formulas in application to our problem. How to Describe a Resulting Expert Estimate For \Deltay? Extension ... |

108 |
Operations on fuzzy numbers
- Dubois, Prade
- 1978
(Show Context)
Citation Context ...perties,s0 is non-decreasing for e ! 0, and it is nonincreasing for e ? 0. For such functionss0 , membership functionss0 (e=oe) were called LR-numbers by D. Dubois and H. Prade who introduced them in =-=[17, 18]. They als-=-o described how to apply arithmetic operations to these "fuzzy numbers". For detailed expositions of LR-numbers, see [19], Section II.2.B; [30], Section 1.9; [20], Chapter 2 and Section 3.2.... |

99 |
Uncertain Dynamic Systems
- Schweppe
- 1973
(Show Context)
Citation Context ...Special Case When the Set G is an Ellipsoid. An important particular case is when the set G is an ellipsoid with a center at ~ 0. Ellipsoids are often used to describe uncertainty domains; see, e.g., =-=[4, 11, 12, 22, 24, 51, 57, 58, 60, 61]-=-. Moreover, it has been shown, both experimentally [11, 12] and theoretically [23], that for many practical problems, ellipsoids are an optimal family for describing domains. For ellipsoids, we can de... |

90 |
Applied numerical analysis
- Gerald, Wheatley
- 1989
(Show Context)
Citation Context ...rease in computation time. Comment. The above numerical differentiation algorithm is the simplest possible. There are other algorithms that give better results for non-linear functions 10 (see, e.g., =-=[26]-=-); e.g., we can estimate c i as f ;i = 1 h i \Delta ` f ` ~ x 1 ; : : : ; ~ x i\Gamma1 ; ~ x i + h i 2 ; ~ x i+1 ; : : : ; ~ x n ' \Gamma f ` ~ x 1 ; : : : ; ~ x i\Gamma1 ; ~ x i \Gamma h i 2 ; ~ x i+... |

77 |
Applications of continuous mathematics in computer science
- Nguyen, Kreinovich
- 1997
(Show Context)
Citation Context ...s the algorithm described in Theorem 1. Comment. For a general overview of how the general ideas of symmetry and invariance help in solving optimization problems in computer-related areas, see, e.g., =-=[50]-=-. When Looking for an Optimal Algorithm, We Can Always Assume That the Vectors ~ ffi (k) Produced by an Algorithm Form an Orthonormal Basis. Indeed, let us show that an arbitrary (n; N)-algorithm can ... |

75 |
Recursive state estimation: Unknown but bounded errors and system inputs
- Schweppe
- 1968
(Show Context)
Citation Context ...Special Case When the Set G is an Ellipsoid. An important particular case is when the set G is an ellipsoid with a center at ~ 0. Ellipsoids are often used to describe uncertainty domains; see, e.g., =-=[4, 11, 12, 22, 24, 51, 57, 58, 60, 61]-=-. Moreover, it has been shown, both experimentally [11, 12] and theoretically [23], that for many practical problems, ellipsoids are an optimal family for describing domains. For ellipsoids, we can de... |

56 |
Measurement errors: theory and practice
- Rabinovich
- 1993
(Show Context)
Citation Context ... only know intervals will be one of the settings considered in this paper. 4 Another setting which we will consider is a setting described in standard engineering textbooks on measurement (see, e.g., =-=[25, 56]-=-; see also [13, 28]). In this setting, the measurement error \Deltax i of each direct measurement is normally distributed with 0 average and known standard deviation oe i , and measurement errors of d... |

50 |
Fuzzy Set Theory –and its Applications
- Zimmerman
- 1991
(Show Context)
Citation Context ...hey also described how to apply arithmetic operations to these "fuzzy numbers". For detailed expositions of LR-numbers, see [19], Section II.2.B; [30], Section 1.9; [20], Chapter 2 and Secti=-=on 3.2.4; [69]-=-, Chapter 5. Let's describe their main formulas in application to our problem. How to Describe a Resulting Expert Estimate For \Deltay? Extension Principle For LR-Numbers: General Idea. For each i, we... |

41 |
On the value of information in system identification - bounded noise case
- Fogel, Huang
- 1982
(Show Context)
Citation Context ...Special Case When the Set G is an Ellipsoid. An important particular case is when the set G is an ellipsoid with a center at ~ 0. Ellipsoids are often used to describe uncertainty domains; see, e.g., =-=[4, 11, 12, 22, 24, 51, 57, 58, 60, 61]-=-. Moreover, it has been shown, both experimentally [11, 12] and theoretically [23], that for many practical problems, ellipsoids are an optimal family for describing domains. For ellipsoids, we can de... |

37 |
State Estimation for Dynamic Systems
- Chernousko
- 1994
(Show Context)
Citation Context |

26 | error estimation, and other applications of set-valued analysis
- Finkelstein, Kosheleva, et al.
(Show Context)
Citation Context ...nter at ~ 0. Ellipsoids are often used to describe uncertainty domains; see, e.g., [4, 11, 12, 22, 24, 51, 57, 58, 60, 61]. Moreover, it has been shown, both experimentally [11, 12] and theoretically =-=[23]-=-, that for many practical problems, ellipsoids are an optimal family for describing domains. For ellipsoids, we can describe the optimal error estimation algorithm (the idea of this algorithm was firs... |

21 |
Integration of fuzzy logical information
- Oden
- 1977
(Show Context)
Citation Context ...nd "or" to the component ones A and B) if we know the degrees of beliefs t(A) and t(B) of the component statements. Which &- and -Operations Should We Choose. Experiments with expert estimat=-=es showed [29, 53, 68]-=- that, on average, the following estimates describe the human reasoning the best: min(t(A); t(B)) for t(A&B) and max(t(A); t(B)) for t(AsB). Second best fits are t(A) \Delta t(B) for t(A&B) and t(A) +... |

19 |
Gerstenberger Estimating uncertainties for geophysical tomography, Reliable Computing
- Doser, Crain, et al.
- 1998
(Show Context)
Citation Context ...for the above optimal algorithm (and for all other algorithms described above), the vectorssdo not depend on the results of the previous calls. Thus, we can make all the calls in parallel (see, e.g., =-=[6, 7, 9, 14, 15, 16, 39, 40, 41, 42, 47, 63]-=-). If we do not have the actual parallel machine, we can use several processors connected into a net [7, 9, 47]. The more processors we have, the less time the resulting computation will take. If we h... |

18 |
Fuzzy sets", Information and control
- Zadeh
- 1965
(Show Context)
Citation Context ...y numbers from the interval [0,1] (0 corresponds to no belief, 1 to absolute belief, and values between 0 and 1 correspond to intermediate degrees of belief). This idea was first proposed by L. Zadeh =-=[66]-=- under the name of fuzzy set theory. There are many different procedures that enable to obtain these numerical values (see, e.g., [19, 32]). If we apply any of these procedures, then for every e, we g... |

17 |
Identification and application of bounded parameter models
- Norton
- 1987
(Show Context)
Citation Context |

16 |
Estimating the Measurement Errors
- Novitskii, Zograph
- 1991
(Show Context)
Citation Context ... indeed close to normal. The second reason for choosing Gaussian distribution is empirical: for the majority of measuring instruments, the measurement error is indeed normally distributed (see, e.g., =-=[52, 54]-=-). In This Paper, WeWill Only Consider Situations When the Measurements Are Reasonably Accurate. In this paper, we will only consider situations in which the direct measurements are accurate enough, s... |

12 |
An improved parameter identification algorithm for signal with unknown-but-bounded errors
- Belfonte, Bona
- 1985
(Show Context)
Citation Context |

12 |
Parallel computers estimate errors caused by imprecise data
- Kreinovich, Bernat, et al.
- 1991
(Show Context)
Citation Context ...mating oe (which is better than the Monte-Carlo simulation), and a new algorithm for estimating \Delta. Our main ideas and the preliminary (restricted) versions of our results were first announced in =-=[33, 37, 40, 41, 42, 44, 45]-=-. 2. ERROR ESTIMATION FOR INDIRECT MEASUREMENT: STATISTICAL SETTING 2.1 ERROR ESTIMATION FOR INDIRECT MEASUREMENT REFORMULATED AS A TOMOGRAPHY PROBLEM When oe 1 = : : : = oe n = 1, The Error Estimatio... |

9 |
Error estimate of the result of indirect measurements by using a calculational experiment, Measurement Techniques
- Kreinovich, Pavlovich
- 1985
(Show Context)
Citation Context ...mating oe (which is better than the Monte-Carlo simulation), and a new algorithm for estimating \Delta. Our main ideas and the preliminary (restricted) versions of our results were first announced in =-=[33, 37, 40, 41, 42, 44, 45]-=-. 2. ERROR ESTIMATION FOR INDIRECT MEASUREMENT: STATISTICAL SETTING 2.1 ERROR ESTIMATION FOR INDIRECT MEASUREMENT REFORMULATED AS A TOMOGRAPHY PROBLEM When oe 1 = : : : = oe n = 1, The Error Estimatio... |

8 |
A fuzzy-set approach to modifiers and vagueness in natural languages
- Hersch, Caramazza
(Show Context)
Citation Context ...nd "or" to the component ones A and B) if we know the degrees of beliefs t(A) and t(B) of the component statements. Which &- and -Operations Should We Choose. Experiments with expert estimat=-=es showed [29, 53, 68]-=- that, on average, the following estimates describe the human reasoning the best: min(t(A); t(B)) for t(A&B) and max(t(A); t(B)) for t(AsB). Second best fits are t(A) \Delta t(B) for t(A&B) and t(A) +... |

8 | Gaussian membership functions are most adequate in representing uncertainty in measurements
- Kreinovich, Quintana, et al.
- 1992
(Show Context)
Citation Context ...d Prade considered a Gaussian membership functions0 (e) = exp(\Gammafi \Delta e 2 ) for some constant fi ? 0 (additional arguments in favor of choosing a Gaussian membership function are presented in =-=[46]-=-). For this function, they showed that the extension principle leads to the membership functions0 (\Deltay=oe), where oe is determined by the formula (3). This is exactly the same formula as for stati... |

8 |
Results of empirical studies in fuzzy set theory
- ZIMMERMAN
- 1978
(Show Context)
Citation Context ...nd "or" to the component ones A and B) if we know the degrees of beliefs t(A) and t(B) of the component statements. Which &- and -Operations Should We Choose. Experiments with expert estimat=-=es showed [29, 53, 68]-=- that, on average, the following estimates describe the human reasoning the best: min(t(A); t(B)) for t(A&B) and max(t(A); t(B)) for t(AsB). Second best fits are t(A) \Delta t(B) for t(A&B) and t(A) +... |

8 |
Orlov, “How often are the observations normal
- I
(Show Context)
Citation Context ... indeed close to normal. The second reason for choosing Gaussian distribution is empirical: for the majority of measuring instruments, the measurement error is indeed normally distributed (see, e.g., =-=[52, 54]-=-). In This Paper, We Will Only Consider Situations When the Measurements Are Reasonably Accurate. In this paper, we will only consider situations in which the direct measurements are accurate enough, ... |

7 | Parallel algorithms for interval computations: an introduction, Interval Computations
- Bernat, Kreinovich
- 1994
(Show Context)
Citation Context ...e above optimal algorithm (and for all other algorithms described above), the vectors ~ ffi do not depend on the results of the previous calls. Thus, we can make all the calls in parallel (see, e.g., =-=[6, 7, 9, 14, 15, 16, 39, 40, 41, 42, 47, 63]-=-). If we do not have the actual parallel machine, we can use several processors connected into a net [7, 9, 47]. The more processors we have, the less time the resulting computation will take. If we h... |

7 |
Concurrent engineering and design for manufacture for electronics products
- Shina
(Show Context)
Citation Context ...ign. So, we are only interested in the interval of possible deviations \Deltay = e y \Gamma y of this characteristic from the nominal one. Computation of such an interval is called tolerance analysis =-=[2, 3, 55, 59, 64]-=-. Tolerances are small, so we can usually safely neglect quadratic terms and consider a linearized problem. From the physical viewpoint, this situation is slightly different from our main problem: we ... |

6 |
Kearfott and V. Kreinovich (eds.), Applications of Interval Computations
- B
- 1996
(Show Context)
Citation Context ...s on the measurement errors, the problem of estimating the error of indirect measurement is called the problem of interval computations; for details and examples of practical applications, see, e.g., =-=[31, 65]-=-. The setting when we only know intervals will be one of the settings considered in this paper. 4 Another setting which we will consider is a setting described in standard engineering textbooks on mea... |

5 |
Estimation of the Phase Space of Dynamic Systems, Nauka publ
- Chernousko
- 1988
(Show Context)
Citation Context |

5 |
On the ellipsoid method for a system of linear differential equations
- Utyubaev
- 1990
(Show Context)
Citation Context |

5 |
Filippov, “Ellipsoidal estimates for a solution of a system of differential equations
- F
- 1992
(Show Context)
Citation Context ...ion. Special Case When the Setsis an Ellipsoid. An important particular case is when the setsis an ellipsoid with a center ats. � Ellipsoids are often used to describe uncertainty domains; see, e.g., =-=[4, 11, 12, 22, 24, 51, 57, 58, 60, 61]-=-. Moreover, it has been shown, both experimentally [11, 12] and theoretically [23], that for many practical problems, ellipsoids are an optimal family for describing domains. For ellipsoids, we can de... |

4 |
Multivariate Error Analysis
- Clifford
- 1973
(Show Context)
Citation Context ...s will be one of the settings considered in this paper. 4 Another setting which we will consider is a setting described in standard engineering textbooks on measurement (see, e.g., [25, 56]; see also =-=[13, 28]-=-). In this setting, the measurement error \Deltax i of each direct measurement is normally distributed with 0 average and known standard deviation oe i , and measurement errors of different direct mea... |

4 |
V.: An optimal method for estimating a random error component
- Koltik, Dmitriev, et al.
- 1986
(Show Context)
Citation Context ...mating oe (which is better than the Monte-Carlo simulation), and a new algorithm for estimating \Delta. Our main ideas and the preliminary (restricted) versions of our results were first announced in =-=[33, 37, 40, 41, 42, 44, 45]-=-. 2. ERROR ESTIMATION FOR INDIRECT MEASUREMENT: STATISTICAL SETTING 2.1 ERROR ESTIMATION FOR INDIRECT MEASUREMENT REFORMULATED AS A TOMOGRAPHY PROBLEM When oe 1 = : : : = oe n = 1, The Error Estimatio... |

4 |
Asymptotic of the function of the outer estimation ellipsoid for a linear singularly perturbed controlled system
- Soltanov
- 1990
(Show Context)
Citation Context |

3 |
Parallel interval computations as a background problem: when processors come and go
- Bernat, Villa, et al.
- 1994
(Show Context)
Citation Context ...e above optimal algorithm (and for all other algorithms described above), the vectors ~ ffi do not depend on the results of the previous calls. Thus, we can make all the calls in parallel (see, e.g., =-=[6, 7, 9, 14, 15, 16, 39, 40, 41, 42, 47, 63]-=-). If we do not have the actual parallel machine, we can use several processors connected into a net [7, 9, 47]. The more processors we have, the less time the resulting computation will take. If we h... |

3 |
PVM estimates errors caused by imprecise data
- Bhamidipati
(Show Context)
Citation Context ...e above optimal algorithm (and for all other algorithms described above), the vectors ~ ffi do not depend on the results of the previous calls. Thus, we can make all the calls in parallel (see, e.g., =-=[6, 7, 9, 14, 15, 16, 39, 40, 41, 42, 47, 63]-=-). If we do not have the actual parallel machine, we can use several processors connected into a net [7, 9, 47]. The more processors we have, the less time the resulting computation will take. If we h... |

3 | A General Approach to Analysis of Uncertainty in Measurements - Kreinovich - 1986 |

3 | A simplified version of the tomography problem can help to estimate the errors of indirect measurements
- Kreinovich
- 1998
(Show Context)
Citation Context |

3 |
The worse, the better: a survey of paradoxical computational complexity of interval computations
- Nesterov, Kreinovich
- 1996
(Show Context)
Citation Context ...osophical Comment: Sometimes, Distortion of Simulated Phenomenon Makes Simulation More Efficient. The use of Cauchy distribution in the above algorithm may seem somewhat counter-intuitive (see, e.g., =-=[36, 49]-=-). Indeed, in the interval setting, we do not know the exact probability distribution of each error \Delta i , but we do know that each error \Delta i belongs to the corresponding interval [\Gamma\Del... |

2 |
How To Find Input Variables Whose Influence On The Result Is The Largest, or, How To Detect Defective Stages
- Beltran, Kreinovich
- 1995
(Show Context)
Citation Context ...ter. However, if there are many parameters (as, e.g., in VLSI manufacturing), then this algorithm takes too long. It turns out that the ideas of Cauchy-based simulation can lead to a faster algorithm =-=[5]-=-. 6.2 PROCESSING EXPERT ESTIMATES The Need for Expert Estimates. In the interval setting, the above algorithms enable us to compute the upper bound \Delta on the error \Deltay of indirect measurement.... |

2 | Intelligent parallel simulation – a key to intractable problems of information processing
- Bernat, Cortes, et al.
- 1992
(Show Context)
Citation Context |

2 |
How to design an expert system that for a given query Q, computes the interval of possible values of probability p(Q) that Q is true, Abstracts for a Workshop on
- Cortes
(Show Context)
Citation Context |

2 |
Fuzzy real algebra: some results", Fuzzy Sets and Systems
- Dubois, Prade
- 1979
(Show Context)
Citation Context ...perties,s0 is non-decreasing for e ! 0, and it is nonincreasing for e ? 0. For such functionss0 , membership functionss0 (e=oe) were called LR-numbers by D. Dubois and H. Prade who introduced them in =-=[17, 18]. They als-=-o described how to apply arithmetic operations to these "fuzzy numbers". For detailed expositions of LR-numbers, see [19], Section II.2.B; [30], Section 1.9; [20], Chapter 2 and Section 3.2.... |

2 |
Filippov, "Ellipsoidal estimates for a solution of a system of differential equations
- F
- 1992
(Show Context)
Citation Context |