and its application in uq
play

and its application in UQ Wenyu Li Arun Hegde Jim Oreluk Andrew - PowerPoint PPT Presentation

Uniform sampling of a feasible set and its application in UQ Wenyu Li Arun Hegde Jim Oreluk Andrew Packard Michael Frenklach SIAM NC17 SPRING 2017 Bound-to-Bound Data Collaboration (B2BDC) Model: Prior Uncertainty Data n Data n Data n


  1. Uniform sampling of a feasible set and its application in UQ Wenyu Li Arun Hegde Jim Oreluk Andrew Packard Michael Frenklach SIAM NC17 SPRING 2017

  2. Bound-to-Bound Data Collaboration (B2BDC) Model: Prior Uncertainty Data n Data n Data n Data 3 Data 2 Data 1 Feasible set SIAM NC17 SPRING 2017

  3. Uniform sampling Goal: uniform sampling of feasible set • Sampling is useful in providing information about • B2BDC makes NO distribution assumptions, but as far as taking samples, uniform distribution of is reasonable • Applying Bayesian analysis with specific prior assumptions also leads to uniform distribution of as posterior (shown in next slide) SIAM NC17 SPRING 2017

  4. What Bayesian analysis leads to Deterministic model: Measurement distribution Prior distribution Posterior distribution Bayesian analysis SIAM NC17 SPRING 2017

  5. B2BDC and Bayesian Calibration and Prediction (BCP) Reference [1] Frenklach, M., Packard, A., Garcia-Donato, G., Paulo, R. and Sacks, J., 2016. Comparison of Statistical and Deterministic Frameworks of Uncertainty Quantification. SIAM/ASA Journal on Uncertainty Quantification , 4 (1), pp.875-901. Nomenclature • sampling efficiency acceptance rate • feasible set SIAM NC17 SPRING 2017

  6. Rejection sampling with box Procedure: • find a bounding box “B2B” Box - available from B2BDC Circumscribed box • generate uniformly distributed samples in the box as candidates • reject the points outside of feasible set Pros & Cons • provably uniform in the feasible set • practical in low dimensions Feasible set • impractical in higher dimensions SIAM NC17 SPRING 2017

  7. Random walk (RW) Procedure: Feasible set • start from a feasible point - available from B2BDC Extreme point • select a random direction, calculate extreme New moving points and choose the next point uniformly direction Moving Next point • repeat the process direction Starting Pros & Cons point • NOT limited by problem dimensions Extreme point • NOT necessarily uniform in the feasible set SIAM NC17 SPRING 2017

  8. Rejection sampling with polytope Feasible set Procedure: Circumscribed polytope • find a bounding polytope • generate candidate points by random walk • reject the points outside of feasible set 6 facets Pros & Cons • provably uniform in the feasible set • increased efficiency with more polytope facets 8 facets SIAM NC17 SPRING 2017

  9. Rejection sampling with polytope Feasible set Procedure: Circumscribed polytope • find a bounding polytope • generate candidate points by random walk • reject the points outside of feasible set 6 facets Pros & Cons • provably uniform in the feasible set • increased efficiency with more polytope facets • practical in low to medium dimensions • limited by computational resource 10 facets SIAM NC17 SPRING 2017

  10. Approximation strategy Procedure: • relax the requirement that the polytope needs Feasible set to contain the feasible set completely Approximate • generate candidate points by random walk polytope • reject the points outside of feasible set Pros & Cons • practical in medium to high dimensions • samples don’t cover the whole feasible set SIAM NC17 SPRING 2017

  11. Define the polytope: one facet Inner and Outer bounds from B2B prediction • Outer bound from optimization ( NO approximation, provably uniform ) • Inner bound from optimization (less aggressive approximation, very close to circumscribed bound) • Sample bound (more aggressive approximation, performance depends on problem) Sample bound from random walk SIAM NC17 SPRING 2017

  12. Effect on sampling efficiency Efficiency density function Projected area Condition for improved efficiency SIAM NC17 SPRING 2017

  13. Effect on sampling efficiency Special case with bounding box Assumption in the polytope case Posterior check SIAM NC17 SPRING 2017

  14. Effect on sampled distribution Target distribution Approximated distribution Difference of mean for a function SIAM NC17 SPRING 2017

  15. Toy example Posterior check Test condition: • 5 parameters, 30 constraints • 1000 facets for each polytope Outer -> Inner : 1.33 > 0.68 • Optimization and sample bounds • Inner -> Sample : 1.40 > 1.33 1000 sample points Polytope Efficiency bound (%) Outer bound 0.095 Inner bound 20.8 Sample bound 27.7 SIAM NC17 SPRING 2017

  16. Toy example Test condition: • 5 parameters, 30 constraints • 1000 facets for each polytope • Optimization and sample bounds • 1000 sample points Polytope Efficiency bound (%) Passed the Kolmogorov-Smirnov Outer bound 0.095 test with 0.05 significance level Inner bound 20.8 Sample bound 27.7 SIAM NC17 SPRING 2017

  17. Principal component analysis (PCA) Feasible set Procedure: • collect RW samples from the feasible set • conduct PCA on RW samples • find a subspace based on PCA result • generate uniform samples in the subspace Pros & Cons • reduced problem dimension • works only if feasible set approximates lower-dimensional manifold/subspace Lower-dimensional subspace SIAM NC17 SPRING 2017

  18. GRI-Mech Test condition: • 102 parameters • 76 experimental data Sampling Efficiency 10 7 RW samples for PCA • • 10-65 subspace dimension - 10 4 facets for each polytope - 10 7 candidate points for sampling Test methods: • polytope and box • inner and sample bounds Subspace dimension SIAM NC17 SPRING 2017

  19. GRI-Mech: 1-D posterior marginal uncertainty Outer bound Inner bound Uniform histogram Test condition: • 45 subspace dimension • Polytope with sample bound 10 4 facets for the polytope • • 1000 sample points • [-1, 1] are prior uncertainties SIAM NC17 SPRING 2017

  20. GRI-Mech: 2-D posterior joint uncertainty Plots: • 2-D projection • [-1 1] are prior uncertainties • Correlations observed SIAM NC17 SPRING 2017

  21. Summary • We developed methods to generate uniformly distributed samples of a feasible set • Approximation strategy and PCA further improves the practicality of rejection sampling method • Hybrid statistical-deterministic uncertainty quantification process combining B2BDC prediction and uniform sampling SIAM NC17 SPRING 2017

  22. Acknowledgements This work is supported as a part of the CCMSC at the University of Utah, funded through PSAAP by the National Nuclear Security Administration, under Award Number DE-NA0002375. SIAM NC17 SPRING 2017

  23. Thank you Questions? SIAM NC17 SPRING 2017

  24. SIAM NC17 SPRING 2017

  25. GRI-Mech: 1-D posterior marginal uncertainty Outer bound Inner bound Uniform sampling, B2BDC Gaussian prior, MCMC Bayes Test condition: • 45 subspace dimension • Polytope with sample bound 10 4 facets for the polytope • • 1000 sample points • [-1, 1] are prior uncertainties SIAM NC17 SPRING 2017

  26. GRI-Mech: 1-D posterior marginal uncertainty Outer bound Inner bound Sample histogram Test condition: • 45 subspace dimension • Polytope with sample bound 10 4 facets for the polytope • True bounds • 1000 sample points • [-1, 1] are prior uncertainties SIAM NC17 SPRING 2017

  27. GRI-Mech: 1-D posterior marginal uncertainty Outer bound Inner bound Uniform histogram Gaussian histogram Test condition: • 45 subspace dimension • Polytope with sample bound 10 4 facets for the polytope • • 1000 sample points • [-1, 1] are prior uncertainties SIAM NC17 SPRING 2017

  28. Rejection sampling with box Procedure: • find a bounding box “B2B” Box - available from B2B Bounding Box • generate uniformly distributed samples in the box as candidates “B2B” box with • reject the points outside of feasible set increased problem dimension Pros & Cons • provably uniform in the feasible set Feasible set • practical in low dimensions SIAM NC17 SPRING 2017

  29. Rejection sampling with polytope Feasible set Procedure: Circumscribed polytope • find a bounding polytope • generate candidate points by random walk • reject the points outside of feasible set Possible Convergence 6 facets Pros & Cons to the convex hull • provably uniform in the feasible set • increased efficiency with more polytope facets • practical in low to medium dimensions 10 facets SIAM NC17 SPRING 2017

  30. Conclusion • Polytope method is in general more practical than box method Sampling Efficiency Box method Polytope method • Approximation method further improves the practicality • PCA and dimension reduction increases efficiency significantly when applicable • Samples of the feasible set provide extra information on posterior uncertainty Subspace dimension SIAM NC17 SPRING 2017

Recommend


More recommend