Granted Patent B2

US 12,724,939 · App. 17/810,257 · Granted Sep 1, 2026

Inventors: Bradley Curtis Lackey (Redmond, WA); Andrew John McGuinness (Thornhill, GB); Yakoub Hassanov Yakoubov (Reigate, GB)

Assignee: Microsoft Technology Licensing, LLC

G06F30/20G06F17/18G06F2111/10

Abstract

Claims (75)

1 . A computing system, comprising:

a processor configured to

receive, for a plurality of correlated variables,

a first predetermined number of simulations from a Monte-Carlo simulation sample, each simulation including a plurality of initial simulation results for the plurality of the variables,

one or more target statistics,

a second predetermined number of strata for each variable and each target statistic, and a cumulative distribution function for each variable and each target statistic;

for each variable and for each of the one or more target statistics,

segment a unit interval of the cumulative distribution function into the second predetermined number of strata and a support of the cumulative distribution function into a plurality of bins such that each bin of the plurality of bins corresponds to one of the strata,

determine an initial discrepancy score based upon a quantity of values in each bin, the first predetermined number of the simulations, and the second predetermined number of the strata for the variable;

determine an initial sum of the initial discrepancy scores;

remove at least one of the plurality of the initial simulations based upon a determination that the initial sum of the initial discrepancy scores is not within an optimization threshold to decrease a sampling error;

add at least one other simulation to a remaining one or more initial simulations based on a policy to minimize the initial discrepancy score;

for each variable and for each of the one or more target statistics, use the quantity of the values in one or more bins corresponding to the at least one of the plurality of the initial simulations and the quantity of the values in one or more bins corresponding to the at least one other simulations to generate an updated discrepancy score;

determine an updated sum of the updated discrepancy scores; and

output a plurality of representative simulations that represent the cumulative distribution function across the strata based upon the updated sum of the updated discrepancy scores.

2 . The computing system of claim 1 , wherein the processor is further configured to accept or reject the at least one other simulation based upon the updated sum of the updated discrepancy scores.

3 . The computing system of claim 1 , wherein the processor is further configured to, for each variable and for each of the one or more target statistics, determine a stratum of a value for each initial simulation result and place the value into one of the plurality of bins based upon the stratum.

4 . The computing system of claim 1 , wherein the at least one other simulation includes a precomputed simulation.

5 . The computing system of claim 1 , wherein the processor is further configured to determine an initial bin-wise discrepancy metric for each bin of the plurality of bins.

6 . The computing system of claim 5 , wherein the initial bin-wise discrepancy metric for a selected bin includes a difference between the quantity of values in the selected bin and the first predetermined number of simulations divided by the second predetermined number of the strata for the variable or target statistic.

7 . The computing system of claim 5 , wherein the initial discrepancy score comprises a maximum bin-wise discrepancy metric or a sum of the initial bin-wise discrepancy metrics for the plurality of bins.

8 . The computing system of claim 5 , wherein the processor is further configured to weight the initial bin-wise discrepancy metric for each bin of the plurality of bins based upon a proximity of a stratum corresponding to the bin to a tail of the cumulative distribution function.

9 . The computing system of claim 5 , wherein the processor is further configured to:

determine an updated bin-wise discrepancy metric for the one or more bins corresponding to the at least one of the plurality of the initial simulations and for the one or more bins corresponding to the at least one other simulation; and

use the updated bin-wise discrepancy metric for the one or more bins corresponding to the at least one of the plurality of the initial simulations and for the one or more bins corresponding to the at least one other simulation, and the initial bin-wise discrepancy metric for each of the remaining one or more initial simulations, to determine the updated discrepancy score.

10 . The computing system of claim 1 , wherein the processor is further configured to, for each variable and target statistic:

decrement the quantity of values in the one or more bins corresponding to the at least one of the plurality of the initial simulations; and

increment the quantity of values in the one or more bins corresponding to the at least one other simulation.

11 . At a computing device, a method comprising:

receiving, for a plurality of correlated variables,

a first predetermined number of simulations from a Monte-Carlo simulation sample, each simulation including a plurality of initial simulation results for the plurality of the variables,

one or more target statistics,

a second predetermined number of strata for each variable and each target statistic, and

a cumulative distribution function for each variable and each target statistic;

for each variable and for each of the one or more target statistics,

segmenting a unit interval of the cumulative distribution function into the second predetermined number of the strata and a support of the cumulative distribution function into a plurality of bins such that each bin of the plurality of bins corresponds to one of the strata,

determining an initial discrepancy score based upon a quantity of values in each bin, the first predetermined number of the simulations, and the second predetermined number of the strata for the variable,

determining an initial sum of the initial discrepancy scores;

removing at least one of the plurality of the initial simulations based upon a determination that the initial sum of the initial discrepancy scores is not within an optimization threshold to decrease a sampling error;

adding at least one other simulation to a remaining one or more initial simulation based on a policy to minimize the initial discrepancy score;

for each variable and for each of the one or more target statistics, using the quantity of the values in one or more bins corresponding to the at least one of the plurality of the initial simulation results and the quantity of the values in one or more bins corresponding to the at least one other simulation result to generate an updated discrepancy score;

determining an updated sum of the updated discrepancy scores; and

outputting a plurality of representative simulations that represent the cumulative distribution function across the strata based upon the updated sum of the updated discrepancy scores.

12 . The method of claim 11 , further comprising accepting or rejecting the at least one other simulation based upon the updated sum of the updated discrepancy scores.

13 . The method of claim 11 , wherein the at least one other simulation includes a precomputed simulation.

14 . The method of claim 11 , wherein determining the initial discrepancy score includes determining an initial bin-wise discrepancy metric for each bin of the plurality of bins.

15 . The method of claim 14 , wherein determining the initial bin-wise discrepancy metric for a selected bin includes determining a difference between the quantity of values in the selected bin and the first predetermined number of the simulations divided by the second predetermined number of the strata for the variable or target statistic.

16 . The method of claim 14 , wherein determining the initial bin-wise discrepancy metric for the selected bin includes determining a maximum bin-wise discrepancy metric or determining a sum of the initial bin-wise discrepancy metrics for the plurality of bins.

17 . The method of claim 14 , wherein determining the initial discrepancy score includes weighting the initial bin-wise discrepancy metric for each bin of the plurality of bins based upon a proximity of a stratum corresponding to the bin to a tail of the cumulative distribution function.

18 . The method of claim 14 , wherein generating the updated discrepancy score includes:

determining an updated bin-wise discrepancy metric for the one or more bins corresponding to the at least one of the plurality of the initial simulation results and for the one or more bins corresponding to the at least one other simulation; and

using the updated bin-wise discrepancy metric for the one or more bins corresponding to the at least one of the plurality of the initial simulation results and for the one or more bins corresponding to the at least one other simulation result, and the initial bin-wise discrepancy metric for each of the remaining one or more initial simulation results, to determine the updated discrepancy score.

19 . The method of claim 11 , wherein generating the updated discrepancy score includes:

decrementing the quantity of values in the one or more bins corresponding to the at least one of the plurality of the initial simulations; and

incrementing the quantity of values in the one or more bins corresponding to the at least one other simulation.

20 . A computing system, comprising:

a processor configured to,

receive, for a plurality of correlated variables,

a first predetermined number of simulations from a Monte-Carlo simulation sample, each simulation including a plurality of initial simulation results for the plurality of the variables;

one or more target statistics,

a second predetermined number of strata for each variable and each target statistic; and

a cumulative distribution function for each variable and each target statistic;

for each variable and for each of the one or more target statistics,

segment a unit interval of the cumulative distribution function into the second predetermined number of strata and a support of the cumulative distribution function into a plurality of bins such that each bin of the plurality of bins corresponds to one of the strata,

count a quantity of values in each bin of the plurality of bins, and

determine an initial discrepancy score based upon a difference between the quantity of values in each bin and the quantity of the initial simulations divided by the second predetermined number of the strata for the variable or target statistic;

determine an initial sum of the initial discrepancy scores;

remove at least one of the plurality of the initial simulation results based upon a determination that the initial sum of the initial discrepancy scores is not within an optimization threshold to decrease a sampling error;

add at least one other simulation to a remaining one or more initial simulations based on a policy to minimize the initial discrepancy score;

for each variable and target statistic,

decrement the quantity of values in one or more bins corresponding to the at least one of the plurality of the initial simulation results,

increment the quantity of values in the one or more bins corresponding to the at least one other simulation result, and

use the quantity of the values in the one or more bins corresponding to the at least one of the plurality of the initial simulation results and the quantity of the values in the one or more bins corresponding to the at least one other simulation result to generate an updated discrepancy score;

determine an updated sum of the updated discrepancy scores; and

output a plurality of representative simulations that represent the cumulative distribution function across the strata based upon the updated sum of the updated discrepancy scores.

Assignments (1)

Continuity (1)

References Cited (32)