Relevance Vector Machines for Earthquake Response Spectra

Size: px
Start display at page:

Download "Relevance Vector Machines for Earthquake Response Spectra"

Transcription

1 American American Transactions Transactions on on Engineering Engineering & Applied Applied Sciences Sciences. American Transactions on Engineering & Applied Sciences Relevance Vector Machines for Earthquake Response Spectra Jale Tezcan a*, Qiang Cheng b a Department of Civil and Environmental Engineering, Southern Illinois University Carbondale, Carbondale, IL 62901, USA b Department of Computer Science, Southern Illinois University Carbondale, Carbondale, IL 62901, USA A R T I C L E I N F O Article history: Received 23 August 2011 Received in revised form 23 September 2011 Accepted 26 September 2011 Available online 26 September 2011 Keywords: Response spectrum Ground motion Supervised learning Bayesian regression Relevance Vector Machines A B S T RA C T This study uses Relevance Vector Machine (RVM) regression to develop a probabilistic model for the average horizontal component of 5%-damped earthquake response spectra. Unlike conventional models, the proposed approach does not require a functional form, and constructs the model based on a set predictive variables and a set of representative ground motion records. The RVM uses Bayesian inference to determine the confidence intervals, instead of estimating them from the mean squared errors on the training set. An example application using three predictive variables (magnitude, distance and fault mechanism) is presented for sites with shear wave velocities ranging from 450 m/s to 900 m/s. The predictions from the proposed model are compared to an existing parametric model. The results demonstrate the validity of the proposed model, and suggest that it can be used as an alternative to the conventional ground motion models. Future studies will investigate the effect of additional predictive variables on the predictive performance of the model American Transactions on Engineering & Applied Sciences. 25

2 1. Introduction Reliable prediction of ground motions from future earthquakes is one of the primary challenges in seismic hazard assessment. Conventional ground motion models are based on parametric regression, which requires a fixed functional form for the predictive model. Because the mechanisms governing ground motion processes are not fully understood, identification of the mathematical form of the underlying function is a challenge. Once a functional form is selected, the model is fit to the data and the model coefficients minimizing the mean squared errors between the model and the data are determined. This approach, when the selected mathematical form does not accurately represent the actual input-output relationship, is susceptible to overfitting. Indeed, using a sufficiently complex model, one can achieve a perfect fit to the training data, regardless of the selected mathematical form. However, a perfect fit to the training data does not indicate the predictive performance of the model for new data. Kernel regression offers a convenient way to perform regression without a fixed parametric form, or any knowledge of the underlying probability distribution. A special form of kernel regression, called the Support Vector Regression (SVR) (Drucker et al., 1997) is characterized by its compact representation and its high generalization performance. In SVR, the training data is first transformed into a high dimensional kernel space, and linear regression is performed on the transformed data. The resulting model is a linear combination of nonlinear kernel functions evaluated at a subset of the training input. Combination weights are determined by minimizing a penalized residual function. The SVR has proved successful in many studies since its introduction in The effectiveness of SVR in ground motion modeling has been recently demonstrated (Tezcan and Cheng, 2011), (Tezcan et al., 2010). A well-known weakness of the SVR is the lack of probabilistic outputs. Although the confidence intervals can be constructed using the mean-squared errors, similar to the approach used in conventional ground motion models, the posterior probabilities, which produce the most reliable estimate of prediction intervals, are not given. The lack of probabilistic outputs in the SVR formulation has motivated the development of a new kernel regression model called Relevance Vector Machine (RVM) (Tipping, 2000) which operates in a Bayesian framework. To overcome the limitations of parametric regression while obtaining probabilistic 26 Jale Tezcan and Qiang Cheng

3 predictions, this paper proposes a new ground motion model based on the RVM regression. Unlike standard ground motion models, which make point estimates of the optimal value of the weights by minimizing the fitting error, the RVM model treats the model coefficients as random variables with independent variances and attempts to find the model that maximizes the likelihood of the observations. This approach offers two main advantages over the conventional ground motion models. First, the prediction uncertainty is explicitly determined using Bayesian inference, as opposed to being estimated from the mean squared errors. Second, the complexity of the RVM model is controlled by assigning suitable prior distributions over the model coefficients, which reduces the overfit susceptibility of the model. The rest of the paper is organized as follows. In Section 2, the RVM regression algorithm is described. Section 3 is devoted to the construction of ground motion model. Starting with the description of the ground motion data and the predictive and target variables, the training results are presented, and the prediction procedure for new data is described. Section 4 demonstrates computational results and compares the RVM predictions to an existing empirical parametric model. Section 5 concludes the paper by presenting the main conclusions of this study, and discusses the advantages and limitations of the proposed method. 2. The RVM Regression Algorithm Given a set of input vectors x i, i = 1: N and corresponding real-valued targets t i, the regression task is to estimate the underlying input-output relationship. Using kernel representation (Smola and Schölkopf, 2004), the regression function can be written as a linear combination of a set of nonlinear kernel functions: N f(x) = w i K(x, x i ) + w 0 i=1 (1) where w i, i = 1 N are the combination weights and w 0 is the bias term. 27

4 This study uses the radial basis function (RBF) kernel: K(x i, x j ), = e γ x i x j 2, γ > 0 (2) where γ is the width parameter controlling the trade-off between model accuracy and complexity. In this study, the width parameter has been determined using cross-validation. Assuming independent noise samples from a zero-mean Gaussian distribution, i.e., n i ~N(0, σ n 2 ), the target values can be written as: t i = f(x i ) + n i i = 1,, N. (3) Recast in matrix from, Equation (3) becomes: t = Φw + n, (4) where t = (t 1,, t N ) T, w = (w 0,, w N ) T, and Φ is an N N + 1 basis matrix with Φ i1 = 1 and Φ ij = K x i, x j 1. The likelihood of the entire set, assuming independent observations is given by: p(t w, σ n 2 ) = (2πσ n 2 ) N 2 e 1 2σ n 2 t Φμ 2. (5) where μ = (μ 0,, μ N ) T is the vector containing the mean values of the combination weights. To control the complexity of the model, a zero-mean Gaussian prior is used where each weight is assigned a different variance (MacKay, 1992): N p(w α) = N(0, 1/α i ). i=0 (6) 28 Jale Tezcan and Qiang Cheng

5 In Eq. (6), α = (α 0,, α N ) where 1/α i is the variance of w i. The posterior distribution of the weights is obtained as: p(w t, α, σ n 2 ) = (2π) N+1 2 C 1 2 e 1 2 (w μ)t C 1 (w μ). (7) where the mean vector μ and covariance matrix C are: μ = σ n 2 C Φ T t (8) C = [σ n 2 Φ T Φ + A ] 1 (9) with α 0 0 : α A = 1. (10) 0 α N The marginal likelihood of the dataset can be determined by integrating out the weights (MacKay, 1992) as follows: p(t α, σ n 2 ) = (2π) N 2 H 1 2 e 1 2 tt H 1 t (11) where H = σ 2 n I N + ΦA 1 Φ T and I N is the identity matrix of size N. Ideal Bayesian inference requires defining prior distributions over α and σ 2 n, followed by marginalization. This process, 2 however, will not result in a closed form solution. Instead, the α i and σ n values maximizing Eq. (11) can be found iteratively as follows (MacKay, 1992): 29

6 (α i ) new = 1 α ic ii μ i 2 (12) (σ n 2 ) new = t Φμ 2 N (1 α i C ii ). (13) Because the nominator in Eq.(12) is a positive number with a maximum value of 1, an α i value tending to infinity implies that the posterior distribution of w i is infinitely peaked at zero, i.e. w i = 0. As a consequence, the corresponding kernel function can be removed from the model. The procedure for determining the weights and the noise variance can be summarized as follows: 1) Select a width parameter of the kernel function and form the basis matrix Φ. 2) Initialize α = (α 0,, α N ) and σ 2 n. 3) Compute matrix A using Eq.(10). 4) Compute the covariance matrix C using Eq.(9). 5) Compute the mean vector μ using Eq.(8). 2 6) Update α and σ n using Eq.(12) and Eq.(13). 7) If α i, set w i = 0 and remove the corresponding column in Φ. 8) Go back to step 3 until convergence. 9) Set the remaining weights equal to μ. The training input points corresponding to the remaining nonzero weights are called the relevance vectors. After the weights and the noise variance are determined, the predictive mean for a new input x can be found as follows: f(x ) = w T Φ. (14) In Eq.(14) Φ = [1 K(x, r 1 ) K(x, r 2 ) K(x, r Nr )] T where relevance vectors. (r 1, r 2, r Nr ) are the 30 Jale Tezcan and Qiang Cheng

7 The total predictive variance can be found by adding the noise variance to the uncertainty due to the variance of the weights, as follows: σ 2 = σ n 2 + Φ T CΦ. (15) 3. Construction of the Ground Motion Model In this section, RVM regression algorithm will be used to construct a ground motion model. In Section 4, the resulting model will be compared to an existing parametric model by Idriss (Idriss, 2008), which will be referred to as I08 model in this paper. To enable a fair comparison, the dataset and the predictive variables of I08 model have been adopted in this study. The RVM algorithm is independent of the size of the predictive variable set; additional variables can be introduced the set of predictive variables can be customized to specific applications. 3.1 Ground Motion Data The ground motion records used in the training have been obtained from the PEER-NGA database (PEER, 2007). Consistent with the I08 model, a total of 942 free-field records have been selected using the following criteria: Shear wave velocity at the top 30 m (V s30 ) ranging from 450 m/s to 900 m/s, Magnitude larger than 4.5, Closest distance between the station and rupture surface (R) less than 200 km. Detailed information regarding these records can be found in the paper by Idriss (Idriss, 2008). 3.2 Predictive and Target Variables The predictive variable set includes moment magnitude (M), natural logarithm of the closest distance between the station and the rupture surface in kilometers (lnr) and fault mechanism (F). Idriss finds that with the shear wave velocity (V s30 ) constrained to 450 m/s- 900 m/s range, it has 31

8 negligible effect on spectral values up to 1 second. Therefore, V s30 was not used as a predictive variable. Following the convention used in I08 model, earthquakes that have been assigned a fault mechanism type 0 and 1 in the PEER database were merged to a single, strike-slip group, while the rest were considered to be representative of reverse events. In the RVM model, strike-slip and reverse earthquakes are assigned F = 1 and F = 1, respectively. The input vector representing i th record has the following form: x i = [M i lnr i F i ]. (16) A set of eight vibration periods (nt = 8) ranging from 0.01 second to 4 seconds was used in the RVM model. The output for the i th record for the vibration period T j is defined as: y i = lns(t j ) for j = 1 to nt. (17) In Equation (17), lns is the natural logarithm of the average horizontal component of 5%- damped pseudo-acceleration response spectrum. The spectral values(s) represent the median value of the geometric mean of the two horizontal components, computed using non-redundant rotations between 0 and 90 degrees (Boore, 2006). 3.3 Training of the RVM Regression Model As a pre-processing step, M and lnr values were linearly scaled to [-1 1] to achieve uniformity between the ranges of the predictive variables. There is no need to scale the fault mechanism identifier (F) as it was already defined to take either -1 or 1. Because kernel functions use Euclidean distances between pairs of input vectors, such scaling will help prevent numerical problems due to large variations between the ranges of the values that variables can take. In the ground motion data used in this study, the ranges of the predictive variables are 4.53 M 7.68, and 0.32 km R km. Therefore, input scaling takes the following form: 32 Jale Tezcan and Qiang Cheng

9 x = 2M 12.21, 2lnR 4.16, F. (18) The optimal value of the kernel width parameter (γ) for each vibration period was determined using 10-fold cross validation (Webb, 2002). In 10-fold cross validation, the training data is randomly partitioned into 10 subsets of equal size; and the model is trained using 9 subsets, and the remaining subset is used to compute the validation error. This process is repeated 10 times, each time with a different validation subset, and the average validation error for a particular γ is computed. By computing the average validation error over a range of possible γ values, the optimal γ with the smallest average validation error is determined. The resulting γ values for each period are listed in Table 1, along with the standard deviation of noise (σ n ), the mean value of the constant term (W 0 ) and the number of relevance vectors. The relevance vectors and the combination weights (W i ) are listed in Table 2. After the RVM models, one for each vibration period, were trained, standardized residuals were computed. Figure 1 shows the distribution of the standardized residuals, corresponding to T=1 second, with respect to M, R and V s30. The residual distribution patterns for other periods were similar, not indicating any systematic bias. Table 1: Kernel width parameter (γ), logarithmic standard deviation of noise (σ n ), mean value of the bias term(w 0 ) and the number of relevance vectors (N r ), for each period. T (sec) γ σ n W o N r

10 Figure 1: Standardized residuals for T=1 second. Table 2: Mean values of the combination weights (W i ) and the relevance vectors (x i ) T=0.01 s. T=0.05 s. i W i r i i W i r i [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] T=0.1 s. T=0.2 s. i W i r i i W i r i [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] 34 Jale Tezcan and Qiang Cheng

11 Table 2 (continued). T=0.5 s. T=1.0 s. i W i r i i W i r i [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] T=2.0 s. T=4.0 s. i W i r i i W i r i [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] 3.4 Prediction Phase After training, the spectral values for a new input vector x = [M, lnr, F ] can be determined as follows: 1. Scale the input to the range [-1 1] using Eq. (18); 2. Construct the basis vector Φ = [1 K(x, r 1 ) K(x, r 2 ) K(x, r Nr )] T using the relevance vectors from Table 2 and the kernel width parameter from Table 1; 3. Determine the median value of lns using Eq.(14); 4. Obtain the standard deviation of the noise from Table 1. Total uncertainty, if needed, can be determined using Eq.(15). 4. Computational Results The RVM model was tested using different magnitude, distance and fault mechanisms, and the results were compared to the I08 model. Figure 2 shows the median spectral acceleration at T=1 35

12 second, along with the 16 th and 84 th percentile values (±σ n bounds) for strike-slip faults, for M=5 (left) and M=7 (right). The circles in the figure show the spectral values from earthquakes with the same fault mechanism and within ±0.25 magnitude units. Figure 3 shows the same information for reverse faults. For periods about 1 second and longer, it was observed that the median estimates from the RVM model were generally lower than those from the I08 model. At very short distances, within ~20 km of the source, RVM estimates were higher for M=7, for both strike-slip and reverse faulting earthquakes. Figure 2: Median ±σ bounds for spectral acceleration at T=1 second, strike-slip faults. Figure 3: Median ±σ bounds for spectral acceleration at T=1 second, reverse faults. 36 Jale Tezcan and Qiang Cheng

13 Figure 4 presents the results for vibration period T=0.2 second, for strike-slip earthquakes. The results for the reverse faulting earthquakes were similar. For shorter vibration periods, and M=7, RVM estimates were lower than those from the I08 model. For M=5, however, RVM predictions equaled or exceed the I08 predictions. Regarding the variation about the median (noise variance), the predictions from the two models were in general agreement for all vibration periods. Figure 4: Median ±σ bounds for spectral acceleration at T=0.2 second, strike-slip faults. 5. Conclusion This paper proposes an RVM-based model for the average horizontal component of earthquake response spectra. Given a set of predictive variable set, and a set of ground motion records, the RVM model predicts the most likely spectral values in addition to its variability. An example application has been presented where the predictions from the RVM model have been compared to an existing, parametric ground motion model. The results demonstrate the validity of the proposed model, and suggest that it can be used as an alternative to the conventional ground motion models. The RVM model offers the following advantages over its conventional counterparts: (1) There is no need to select a fixed functional form. By determining the optimal variances associated with 37

14 the weights, the RVM automatically detects the most plausible model; (2) The resulting RVM model has a simple mathematical structure (weighted average of exponential basis functions), and is based on a small number of samples that carry the most relevant information. Samples that are not well supported by the evidence (as measured by the increase in the marginal likelihood) are automatically pruned. (3) Because the model complexity is controlled during the training stage, the RVM has lower risk of over-fitting. One limitation of the proposed approach is that the resulting model may be difficult to interpret. Because the RVM is not a physical model, it does not allow any user-defined, physical constraints, not allowing extension of the model to scenarios not represented in the training data set. However, in our opinion, this does not constitute a shortcoming, considering that the reliability such practice is questionable in any regression model. Another potential limitation is that the RVM requires a user-defined kernel width parameter, which does not have a very clear intuitive meaning, especially when working with high dimensional input vectors. However, the optimal value of the kernel width parameter can be determined using cross-validation, as has been done in this study. Future studies will investigate the effect of using additional predictive variables on the performance of the model. 6. Acknowledgements This material is based in part upon work supported by the National Science Foundation under Grant Number CMMI References Boore, D.M., J. Watson-Lamprey, and N.A. Abrahamson. (2006). Orientation-independent measures of ground motion. Bulletin of the Seismological Society of America, 96(4A), Bozorgnia, Y. and K. W. Campbell. (2004). The vertical-to-horizontal response spectral ratio and tentative procedures for developing simplified V/H and vertical design spectra. Journal of Earthquake Engineering, 8(2), Campbell, K. W. and Y. Bozorgnia. (2003). Updated Near-Source Ground-Motion (Attenuation) Relations for the Horizontal and Vertical Components of Peak Ground Acceleration and Acceleration Response Spectra. Bulletin of the Seismological Society of America, 93(1), Jale Tezcan and Qiang Cheng

15 Drucker, H., C. J. C. Burges, L. Kaufman, A. Smola and V. Vapnik. (1997). Support vector regression machines, Advances in Neural Information Processing Systems 9, MIT Press. Idriss, I. M. (2008). An NGA empirical model for estimating the horizontal spectral values generated by shallow crustal earthquakes. Earthquake spectra, 24(1), MacKay, D. J. C. (1992). Bayesian interpolation. Neural computation, 4(3), MacKay, D. J. C. (1992). The evidence framework applied to classification networks. Neural Computation, 4(5), PEER. (2007). PEER-NGA Database. Smola, A. J. and B. Schölkopf. (2004). A tutorial on support vector regression. Statistics and Computing, 14(3), Tezcan, J. and Q. Cheng. (2011). A Nonparametric Characterization of Vertical Ground Motion Effects. Earthquake Engineering and Structural Dynamics (in print). Tezcan, J., Q. Cheng and L. Hill. (2010). Response Spectrum Estimation using Support Vector Machines, 5th International Conference on Recent Advances in Geotechnical Earthquake Engineering and Soil Dynamics, San Diego, CA. Tipping, M. (2000). The relevance vector machine. Advances in Neural Information Processing Systems MIT Press. Webb, A. (2002). Statistical pattern recognition, New York, John Wiley and Sons. Dr.Jale Tezcan is an Associate Professor in the Department of Civil and Environmental Engineering at Southern Illinois University Carbondale. She earned her Ph.D. from Rice University, Houston, TX in Dr.Tezcan s research interests include earthquake engineering, material characterization, and numerical methods. Dr.Qiang Cheng is an Assistant Professor in the Department of Computer Science at Southern Illinois University Carbondale. He earned his Ph.D. from the University of Illinois at Urbana Champaign, IL in Dr.Cheng s research interests include pattern recognition, machine learning and signal processing. Peer Review: This article has been internationally peer-reviewed and accepted for publication according to the guidelines given at the journal s website. 39

Maximum Direction to Geometric Mean Spectral Response Ratios using the Relevance Vector Machine

Maximum Direction to Geometric Mean Spectral Response Ratios using the Relevance Vector Machine Maximum Direction to Geometric Mean Spectral Response Ratios using the Relevance Vector Machine Y. Dak Hazirbaba, J. Tezcan, Q. Cheng Southern Illinois University Carbondale, IL, USA SUMMARY: The 2009

More information

Damping Scaling of Response Spectra for Shallow CCCCCCCCCrustalstallPaper Crustal Earthquakes in Active Tectonic Title Line Regions 1 e 2

Damping Scaling of Response Spectra for Shallow CCCCCCCCCrustalstallPaper Crustal Earthquakes in Active Tectonic Title Line Regions 1 e 2 Damping Scaling of Response Spectra for Shallow CCCCCCCCCrustalstallPaper Crustal Earthquakes in Active Tectonic Title Line Regions 1 e 2 S. Rezaeian U.S. Geological Survey, Golden, CO, USA Y. Bozorgnia

More information

GROUND-MOTION SELECTION FOR PEER TRANSPORTATION RESEARCH PROGRAM

GROUND-MOTION SELECTION FOR PEER TRANSPORTATION RESEARCH PROGRAM JOINT CONFERENCE PROCEEDINGS 7th International Conference on Urban Earthquake Engineering (7CUEE) & 5th International Conference on Earthquake Engineering (5ICEE) March 3-5, 2010, Tokyo Institute of Technology,

More information

RECORD OF REVISIONS. Page 2 of 17 GEO. DCPP.TR.14.06, Rev. 0

RECORD OF REVISIONS. Page 2 of 17 GEO. DCPP.TR.14.06, Rev. 0 Page 2 of 17 RECORD OF REVISIONS Rev. No. Reason for Revision Revision Date 0 Initial Report - this work is being tracked under Notification SAPN 50638425-1 8/6/2014 Page 3 of 17 TABLE OF CONTENTS Page

More information

Relevance Vector Machines

Relevance Vector Machines LUT February 21, 2011 Support Vector Machines Model / Regression Marginal Likelihood Regression Relevance vector machines Exercise Support Vector Machines The relevance vector machine (RVM) is a bayesian

More information

ECE521 week 3: 23/26 January 2017

ECE521 week 3: 23/26 January 2017 ECE521 week 3: 23/26 January 2017 Outline Probabilistic interpretation of linear regression - Maximum likelihood estimation (MLE) - Maximum a posteriori (MAP) estimation Bias-variance trade-off Linear

More information

VALIDATION AGAINST NGA EMPIRICAL MODEL OF SIMULATED MOTIONS FOR M7.8 RUPTURE OF SAN ANDREAS FAULT

VALIDATION AGAINST NGA EMPIRICAL MODEL OF SIMULATED MOTIONS FOR M7.8 RUPTURE OF SAN ANDREAS FAULT VALIDATION AGAINST NGA EMPIRICAL MODEL OF SIMULATED MOTIONS FOR M7.8 RUPTURE OF SAN ANDREAS FAULT L.M. Star 1, J. P. Stewart 1, R.W. Graves 2 and K.W. Hudnut 3 1 Department of Civil and Environmental Engineering,

More information

DIRECT HAZARD ANALYSIS OF INELASTIC RESPONSE SPECTRA

DIRECT HAZARD ANALYSIS OF INELASTIC RESPONSE SPECTRA DIRECT HAZARD ANALYSIS OF INELASTIC RESPONSE SPECTRA ABSTRACT Y. Bozorgnia, M. Hachem, and K.W. Campbell Associate Director, PEER, University of California, Berkeley, California, USA Senior Associate,

More information

Vertical to Horizontal (V/H) Ratios for Large Megathrust Subduction Zone Earthquakes

Vertical to Horizontal (V/H) Ratios for Large Megathrust Subduction Zone Earthquakes Vertical to Horizontal (V/H) Ratios for Large Megathrust Subduction Zone Earthquakes N.J. Gregor Consultant, Oakland, California, USA N.A. Abrahamson University of California, Berkeley, USA K.O. Addo BC

More information

Neutron inverse kinetics via Gaussian Processes

Neutron inverse kinetics via Gaussian Processes Neutron inverse kinetics via Gaussian Processes P. Picca Politecnico di Torino, Torino, Italy R. Furfaro University of Arizona, Tucson, Arizona Outline Introduction Review of inverse kinetics techniques

More information

Non-Ergodic Probabilistic Seismic Hazard Analyses

Non-Ergodic Probabilistic Seismic Hazard Analyses Non-Ergodic Probabilistic Seismic Hazard Analyses M.A. Walling Lettis Consultants International, INC N.A. Abrahamson University of California, Berkeley SUMMARY A method is developed that relaxes the ergodic

More information

Introduction to SVM and RVM

Introduction to SVM and RVM Introduction to SVM and RVM Machine Learning Seminar HUS HVL UIB Yushu Li, UIB Overview Support vector machine SVM First introduced by Vapnik, et al. 1992 Several literature and wide applications Relevance

More information

Updating the Chiou and YoungsNGAModel: Regionalization of Anelastic Attenuation

Updating the Chiou and YoungsNGAModel: Regionalization of Anelastic Attenuation Updating the Chiou and YoungsNGAModel: Regionalization of Anelastic Attenuation B. Chiou California Department of Transportation R.R. Youngs AMEC Environment & Infrastructure SUMMARY: (10 pt) Ground motion

More information

Representative ground-motion ensembles for several major earthquake scenarios in New Zealand

Representative ground-motion ensembles for several major earthquake scenarios in New Zealand Representative ground-motion ensembles for several major earthquake scenarios in New Zealand K. Tarbali & B.A. Bradley Department of Civil and Natural Resources Engineering, University of Canterbury, Christchurch.

More information

NGA-Subduction: Development of the Largest Ground Motion Database for Subduction Events

NGA-Subduction: Development of the Largest Ground Motion Database for Subduction Events NGA-Subduction: Development of the Largest Ground Motion Database for Subduction Events Tadahiro Kishida. Ph.D., and Yousef Bozorgnia, Ph.D., P.E. University of California, Berkeley 1 Probabilistic Seismic

More information

7 Ground Motion Models

7 Ground Motion Models 7 Ground Motion Models 7.1 Introduction Ground motion equations are often called attenution relations but they describe much more than just the attenutation of the ground motion; they describe the probability

More information

ACCOUNTING FOR SITE EFFECTS IN PROBABILISTIC SEISMIC HAZARD ANALYSIS: OVERVIEW OF THE SCEC PHASE III REPORT

ACCOUNTING FOR SITE EFFECTS IN PROBABILISTIC SEISMIC HAZARD ANALYSIS: OVERVIEW OF THE SCEC PHASE III REPORT ACCOUNTING FOR SITE EFFECTS IN PROBABILISTIC SEISMIC HAZARD ANALYSIS: OVERVIEW OF THE SCEC PHASE III REPORT Edward H FIELD 1 And SCEC PHASE III WORKING GROUP 2 SUMMARY Probabilistic seismic hazard analysis

More information

ESANN'2001 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), April 2001, D-Facto public., ISBN ,

ESANN'2001 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), April 2001, D-Facto public., ISBN , Sparse Kernel Canonical Correlation Analysis Lili Tan and Colin Fyfe 2, Λ. Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong. 2. School of Information and Communication

More information

INTRODUCTION TO PATTERN RECOGNITION

INTRODUCTION TO PATTERN RECOGNITION INTRODUCTION TO PATTERN RECOGNITION INSTRUCTOR: WEI DING 1 Pattern Recognition Automatic discovery of regularities in data through the use of computer algorithms With the use of these regularities to take

More information

Advanced Machine Learning Practical 4b Solution: Regression (BLR, GPR & Gradient Boosting)

Advanced Machine Learning Practical 4b Solution: Regression (BLR, GPR & Gradient Boosting) Advanced Machine Learning Practical 4b Solution: Regression (BLR, GPR & Gradient Boosting) Professor: Aude Billard Assistants: Nadia Figueroa, Ilaria Lauzana and Brice Platerrier E-mails: aude.billard@epfl.ch,

More information

Beyond Sa GMRotI : Conversion to Sa Arb, Sa SN, and Sa MaxRot

Beyond Sa GMRotI : Conversion to Sa Arb, Sa SN, and Sa MaxRot Bulletin of the Seismological Society of America, Vol. 97, No. 5, pp. 1511 1524, October 2007, doi: 10.1785/0120070007 Beyond Sa GMRotI : Conversion to Sa Arb, Sa SN, and Sa MaxRot by Jennie A. Watson-Lamprey

More information

Updated Graizer-Kalkan GMPEs (GK13) Southwestern U.S. Ground Motion Characterization SSHAC Level 3 Workshop 2 Berkeley, CA October 23, 2013

Updated Graizer-Kalkan GMPEs (GK13) Southwestern U.S. Ground Motion Characterization SSHAC Level 3 Workshop 2 Berkeley, CA October 23, 2013 Updated Graizer-Kalkan GMPEs (GK13) Southwestern U.S. Ground Motion Characterization SSHAC Level 3 Workshop 2 Berkeley, CA October 23, 2013 PGA Model Our model is based on representation of attenuation

More information

Estimating Earthquake-induced Slope Displacements Using Vector Ground Motion Intensity Measures

Estimating Earthquake-induced Slope Displacements Using Vector Ground Motion Intensity Measures Estimating Earthquake-induced Slope Displacements Using Vector Ground Motion Intensity Measures Gang Wang & Wenqi Du Hong Kong University of Science and Technology, Hong Kong SAR, China SUMMARY: Realistic

More information

Ground-Motion Attenuation Relationships for Subduction- Zone Earthquakes in Northern Taiwan

Ground-Motion Attenuation Relationships for Subduction- Zone Earthquakes in Northern Taiwan Ground-Motion Attenuation Relationships for Subduction- Zone Earthquakes in Northern Taiwan Lin, P.S., Lee, C.T. Bulletin of the Seismology Society of America (2008) Presenter: Yang Pei-Xin Adviser: Lee

More information

Pattern Recognition and Machine Learning. Bishop Chapter 6: Kernel Methods

Pattern Recognition and Machine Learning. Bishop Chapter 6: Kernel Methods Pattern Recognition and Machine Learning Chapter 6: Kernel Methods Vasil Khalidov Alex Kläser December 13, 2007 Training Data: Keep or Discard? Parametric methods (linear/nonlinear) so far: learn parameter

More information

Eleventh U.S. National Conference on Earthquake Engineering Integrating Science, Engineering & Policy June 25-29, 2018 Los Angeles, California

Eleventh U.S. National Conference on Earthquake Engineering Integrating Science, Engineering & Policy June 25-29, 2018 Los Angeles, California Eleventh U.S. National Conference on Earthquake Engineering Integrating Science, Engineering & Policy June 25-29, 2018 Los Angeles, California Site-Specific MCE R Response Spectra for Los Angeles Region

More information

Linear Dependency Between and the Input Noise in -Support Vector Regression

Linear Dependency Between and the Input Noise in -Support Vector Regression 544 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 14, NO. 3, MAY 2003 Linear Dependency Between the Input Noise in -Support Vector Regression James T. Kwok Ivor W. Tsang Abstract In using the -support vector

More information

Hazard Feedback using the. current GMPEs for DCPP. Nick Gregor. PG&E DCPP SSHAC Study. SWUS GMC Workshop 2 October 22, 2013

Hazard Feedback using the. current GMPEs for DCPP. Nick Gregor. PG&E DCPP SSHAC Study. SWUS GMC Workshop 2 October 22, 2013 1 Hazard Feedback using the current GMPEs for DCPP Nick Gregor PG&E DCPP SSHAC Study SWUS GMC Workshop 2 October 22, 2013 PGA Hazard By Source 0.5 Hz 2 Deaggregation AEP = 10-4 PGA 0.5 Hz 3 4 Base Case

More information

Overfitting, Bias / Variance Analysis

Overfitting, Bias / Variance Analysis Overfitting, Bias / Variance Analysis Professor Ameet Talwalkar Professor Ameet Talwalkar CS260 Machine Learning Algorithms February 8, 207 / 40 Outline Administration 2 Review of last lecture 3 Basic

More information

CSC2515 Winter 2015 Introduction to Machine Learning. Lecture 2: Linear regression

CSC2515 Winter 2015 Introduction to Machine Learning. Lecture 2: Linear regression CSC2515 Winter 2015 Introduction to Machine Learning Lecture 2: Linear regression All lecture slides will be available as.pdf on the course website: http://www.cs.toronto.edu/~urtasun/courses/csc2515/csc2515_winter15.html

More information

Ground-Motion Prediction Equations (GMPEs) from a Global Dataset: The PEER NGA Equations

Ground-Motion Prediction Equations (GMPEs) from a Global Dataset: The PEER NGA Equations Ground-Motion Prediction Equations (GMPEs) from a Global Dataset: The PEER NGA Equations David M. Boore U.S. Geological Survey Abstract The PEER NGA ground-motion prediction equations (GMPEs) were derived

More information

Occurrence of negative epsilon in seismic hazard analysis deaggregation, and its impact on target spectra computation

Occurrence of negative epsilon in seismic hazard analysis deaggregation, and its impact on target spectra computation Occurrence of negative epsilon in seismic hazard analysis deaggregation, and its impact on target spectra computation Lynne S. Burks 1 and Jack W. Baker Department of Civil and Environmental Engineering,

More information

Hybrid Empirical Ground-Motion Prediction Equations for Eastern North America Using NGA Models and Updated Seismological Parameters

Hybrid Empirical Ground-Motion Prediction Equations for Eastern North America Using NGA Models and Updated Seismological Parameters Hybrid Empirical Ground-Motion Prediction Equations for Eastern North America Using NGA Models and Updated Seismological Parameters by Shahram Pezeshk, 1 Arash Zandieh, 1 and Behrooz Tavakoli 2 1 Department

More information

Computer Vision Group Prof. Daniel Cremers. 2. Regression (cont.)

Computer Vision Group Prof. Daniel Cremers. 2. Regression (cont.) Prof. Daniel Cremers 2. Regression (cont.) Regression with MLE (Rep.) Assume that y is affected by Gaussian noise : t = f(x, w)+ where Thus, we have p(t x, w, )=N (t; f(x, w), 2 ) 2 Maximum A-Posteriori

More information

Content. Learning. Regression vs Classification. Regression a.k.a. function approximation and Classification a.k.a. pattern recognition

Content. Learning. Regression vs Classification. Regression a.k.a. function approximation and Classification a.k.a. pattern recognition Content Andrew Kusiak Intelligent Systems Laboratory 239 Seamans Center The University of Iowa Iowa City, IA 52242-527 andrew-kusiak@uiowa.edu http://www.icaen.uiowa.edu/~ankusiak Introduction to learning

More information

Linear & nonlinear classifiers

Linear & nonlinear classifiers Linear & nonlinear classifiers Machine Learning Hamid Beigy Sharif University of Technology Fall 1394 Hamid Beigy (Sharif University of Technology) Linear & nonlinear classifiers Fall 1394 1 / 34 Table

More information

Spatial Cross-correlation Models for Vector Intensity Measures (PGA, Ia, PGV and Sa s) Considering Regional Site Conditions

Spatial Cross-correlation Models for Vector Intensity Measures (PGA, Ia, PGV and Sa s) Considering Regional Site Conditions Spatial Cross-correlation Models for Vector Intensity Measures (PGA, Ia, PGV and Sa s) Considering Regional Site Conditions Gang Wang and Wenqi Du Department of Civil and Environmental Engineering Hong

More information

Gaussian Processes (10/16/13)

Gaussian Processes (10/16/13) STA561: Probabilistic machine learning Gaussian Processes (10/16/13) Lecturer: Barbara Engelhardt Scribes: Changwei Hu, Di Jin, Mengdi Wang 1 Introduction In supervised learning, we observe some inputs

More information

Lecture : Probabilistic Machine Learning

Lecture : Probabilistic Machine Learning Lecture : Probabilistic Machine Learning Riashat Islam Reasoning and Learning Lab McGill University September 11, 2018 ML : Many Methods with Many Links Modelling Views of Machine Learning Machine Learning

More information

Southern California ground motion envelopes over ranges of magnitudes, distances, and site conditions

Southern California ground motion envelopes over ranges of magnitudes, distances, and site conditions 55 Chapter 3 Average properties of Southern California ground motion envelopes over ranges of magnitudes, distances, and site conditions In this chapter, I use the envelope attenuation relationships derived

More information

Ground Motion Prediction Equation Hazard Sensitivity Results for Palo Verde Nuclear Generating Station Site (PVNGS)

Ground Motion Prediction Equation Hazard Sensitivity Results for Palo Verde Nuclear Generating Station Site (PVNGS) Ground Motion Prediction Equation Hazard Sensitivity Results for Palo Verde Nuclear Generating Station Site (PVNGS) M.Walling SWUS GMC Workshop 3 March 11, 2013 Hazard ground motion prediction equation

More information

ADVANCED MACHINE LEARNING ADVANCED MACHINE LEARNING. Non-linear regression techniques Part - II

ADVANCED MACHINE LEARNING ADVANCED MACHINE LEARNING. Non-linear regression techniques Part - II 1 Non-linear regression techniques Part - II Regression Algorithms in this Course Support Vector Machine Relevance Vector Machine Support vector regression Boosting random projections Relevance vector

More information

Gaussian Processes. Le Song. Machine Learning II: Advanced Topics CSE 8803ML, Spring 2012

Gaussian Processes. Le Song. Machine Learning II: Advanced Topics CSE 8803ML, Spring 2012 Gaussian Processes Le Song Machine Learning II: Advanced Topics CSE 8803ML, Spring 01 Pictorial view of embedding distribution Transform the entire distribution to expected features Feature space Feature

More information

Seismic hazard modeling for Bulgaria D. Solakov, S. Simeonova

Seismic hazard modeling for Bulgaria D. Solakov, S. Simeonova Seismic hazard modeling for Bulgaria D. Solakov, S. Simeonova Bulgarian seismic network and foreign stations used in epicenter location Sismicity in Bulgaria and surroundings (M 4.) Epicentral map for

More information

Introduction. Chapter 1

Introduction. Chapter 1 Chapter 1 Introduction In this book we will be concerned with supervised learning, which is the problem of learning input-output mappings from empirical data (the training dataset). Depending on the characteristics

More information

Regression. Machine Learning and Pattern Recognition. Chris Williams. School of Informatics, University of Edinburgh.

Regression. Machine Learning and Pattern Recognition. Chris Williams. School of Informatics, University of Edinburgh. Regression Machine Learning and Pattern Recognition Chris Williams School of Informatics, University of Edinburgh September 24 (All of the slides in this course have been adapted from previous versions

More information

The effect of bounds on magnitude, source-to-site distance and site condition in PSHA-based ground motion selection

The effect of bounds on magnitude, source-to-site distance and site condition in PSHA-based ground motion selection The effect of bounds on magnitude, source-to-site distance and site condition in PSHA-based ground motion selection K. Tarbali & B.A. Bradley Department of Civil and Natural Resources Engineering, University

More information

NEXT GENERATION ATTENUATION (NGA) EMPIRICAL GROUND MOTION MODELS: CAN THEY BE USED IN EUROPE?

NEXT GENERATION ATTENUATION (NGA) EMPIRICAL GROUND MOTION MODELS: CAN THEY BE USED IN EUROPE? First European Conference on Earthquake Engineering and Seismology (a joint event of the 13 th ECEE & 30 th General Assembly of the ESC) Geneva, Switzerland, 3-8 September 2006 Paper Number: 458 NEXT GENERATION

More information

The Ranges of Uncertainty among the Use of NGA-West1 and NGA-West 2 Ground Motion Prediction Equations

The Ranges of Uncertainty among the Use of NGA-West1 and NGA-West 2 Ground Motion Prediction Equations The Ranges of Uncertainty among the Use of NGA-West1 and NGA-West 2 Ground otion Prediction Equations T. Ornthammarath Assistant Professor, Department of Civil and Environmental Engineering, Faculty of

More information

Spatial Correlation of Ground Motions in Seismic Hazard Assessment

Spatial Correlation of Ground Motions in Seismic Hazard Assessment Spatial Correlation of Ground Motions in Seismic Hazard Assessment Taojun Liu tliu82@uwo.ca Department of Civil & Environmental Engineering University of Western Ontario London, Ontario, Canada 1 Outline

More information

Hybrid Empirical Ground-Motion Prediction Equations for Eastern North America Using NGA Models and Updated Seismological Parameters

Hybrid Empirical Ground-Motion Prediction Equations for Eastern North America Using NGA Models and Updated Seismological Parameters Bulletin of the Seismological Society of America, Vol. 101, No. 4, pp. 1859 1870, August 2011, doi: 10.1785/0120100144 Hybrid Empirical Ground-Motion Prediction Equations for Eastern North America Using

More information

Overview of Seismic PHSA Approaches with Emphasis on the Management of Uncertainties

Overview of Seismic PHSA Approaches with Emphasis on the Management of Uncertainties H4.SMR/1645-29 "2nd Workshop on Earthquake Engineering for Nuclear Facilities: Uncertainties in Seismic Hazard" 14-25 February 2005 Overview of Seismic PHSA Approaches with Emphasis on the Management of

More information

Relevance of Fault-Normal/Parallel and Maximum Direction Rotated Ground Motions on Nonlinear Behavior of Multi-Story Buildings

Relevance of Fault-Normal/Parallel and Maximum Direction Rotated Ground Motions on Nonlinear Behavior of Multi-Story Buildings Relevance of Fault-Normal/Parallel and Maximum Direction Rotated Ground Motions on Nonlinear Behavior of Multi-Story Buildings J.C. Reyes Universidad de los Andes, Bogotá, Colombia E. Kalkan United States

More information

Statistical Learning Reading Assignments

Statistical Learning Reading Assignments Statistical Learning Reading Assignments S. Gong et al. Dynamic Vision: From Images to Face Recognition, Imperial College Press, 2001 (Chapt. 3, hard copy). T. Evgeniou, M. Pontil, and T. Poggio, "Statistical

More information

Nonparametric Bayesian Methods (Gaussian Processes)

Nonparametric Bayesian Methods (Gaussian Processes) [70240413 Statistical Machine Learning, Spring, 2015] Nonparametric Bayesian Methods (Gaussian Processes) Jun Zhu dcszj@mail.tsinghua.edu.cn http://bigml.cs.tsinghua.edu.cn/~jun State Key Lab of Intelligent

More information

Bayesian Machine Learning

Bayesian Machine Learning Bayesian Machine Learning Andrew Gordon Wilson ORIE 6741 Lecture 2: Bayesian Basics https://people.orie.cornell.edu/andrew/orie6741 Cornell University August 25, 2016 1 / 17 Canonical Machine Learning

More information

Prediction of elastic displacement response spectra in Europe and the Middle East

Prediction of elastic displacement response spectra in Europe and the Middle East EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS Earthquake Engng Struct. Dyn. 2007; 36:1275 1301 Published online 27 February 2007 in Wiley InterScience (www.interscience.wiley.com)..679 Prediction of elastic

More information

Introduction to Gaussian Processes

Introduction to Gaussian Processes Introduction to Gaussian Processes Neil D. Lawrence GPSS 10th June 2013 Book Rasmussen and Williams (2006) Outline The Gaussian Density Covariance from Basis Functions Basis Function Representations Constructing

More information

Development of Ground Motion Time Histories for Seismic Design

Development of Ground Motion Time Histories for Seismic Design Proceedings of the Ninth Pacific Conference on Earthquake Engineering Building an Earthquake-Resilient Society 14-16 April, 2011, Auckland, New Zealand Development of Ground Motion Time Histories for Seismic

More information

NONLINEAR CLASSIFICATION AND REGRESSION. J. Elder CSE 4404/5327 Introduction to Machine Learning and Pattern Recognition

NONLINEAR CLASSIFICATION AND REGRESSION. J. Elder CSE 4404/5327 Introduction to Machine Learning and Pattern Recognition NONLINEAR CLASSIFICATION AND REGRESSION Nonlinear Classification and Regression: Outline 2 Multi-Layer Perceptrons The Back-Propagation Learning Algorithm Generalized Linear Models Radial Basis Function

More information

Probabilistic & Unsupervised Learning

Probabilistic & Unsupervised Learning Probabilistic & Unsupervised Learning Gaussian Processes Maneesh Sahani maneesh@gatsby.ucl.ac.uk Gatsby Computational Neuroscience Unit, and MSc ML/CSML, Dept Computer Science University College London

More information

THE USE OF INPUT ENERGY FOR SEISMIC HAZARD ASSESSMENT WITH DIFFERENT DUCTILITY LEVEL

THE USE OF INPUT ENERGY FOR SEISMIC HAZARD ASSESSMENT WITH DIFFERENT DUCTILITY LEVEL th World Conference on Earthquake Engineering Vancouver, B.C., Canada August -6, Paper No. 8 THE USE OF INPUT ENERGY FOR SEISMIC HAZARD ASSESSMENT WITH DIFFERENT DUCTILITY LEVEL Mao-Sheng GONG And Li-Li

More information

Gaussian Process Regression

Gaussian Process Regression Gaussian Process Regression 4F1 Pattern Recognition, 21 Carl Edward Rasmussen Department of Engineering, University of Cambridge November 11th - 16th, 21 Rasmussen (Engineering, Cambridge) Gaussian Process

More information

PACIFIC EARTHQUAKE ENGINEERING RESEARCH CENTER

PACIFIC EARTHQUAKE ENGINEERING RESEARCH CENTER PACIFIC EARTHQUAKE ENGINEERING RESEARCH CENTER NGA-West2 Models for Ground-Motion Directionality Shrey K. Shahi Jack W. Baker Department of Civil and Environmental Engineering Stanford University PEER

More information

Site specific seismic hazard assessment a case study of Guanyin offshore wind farm 場址特定地震危害度評估 - 以觀音離岸風力發電廠為例

Site specific seismic hazard assessment a case study of Guanyin offshore wind farm 場址特定地震危害度評估 - 以觀音離岸風力發電廠為例 Site specific seismic hazard assessment a case study of Guanyin offshore wind farm 場址特定地震危害度評估 - 以觀音離岸風力發電廠為例 Supervisor : Dr. Chyi-Tyi Lee and Dr. Kuo-Fong Ma Speaker : Jia-Cian Gao 2018/04/26 1 1. A

More information

Machine Learning Lecture 7

Machine Learning Lecture 7 Course Outline Machine Learning Lecture 7 Fundamentals (2 weeks) Bayes Decision Theory Probability Density Estimation Statistical Learning Theory 23.05.2016 Discriminative Approaches (5 weeks) Linear Discriminant

More information

Bayesian Inference: Principles and Practice 3. Sparse Bayesian Models and the Relevance Vector Machine

Bayesian Inference: Principles and Practice 3. Sparse Bayesian Models and the Relevance Vector Machine Bayesian Inference: Principles and Practice 3. Sparse Bayesian Models and the Relevance Vector Machine Mike Tipping Gaussian prior Marginal prior: single α Independent α Cambridge, UK Lecture 3: Overview

More information

Density Estimation. Seungjin Choi

Density Estimation. Seungjin Choi Density Estimation Seungjin Choi Department of Computer Science and Engineering Pohang University of Science and Technology 77 Cheongam-ro, Nam-gu, Pohang 37673, Korea seungjin@postech.ac.kr http://mlg.postech.ac.kr/

More information

Perceptron Revisited: Linear Separators. Support Vector Machines

Perceptron Revisited: Linear Separators. Support Vector Machines Support Vector Machines Perceptron Revisited: Linear Separators Binary classification can be viewed as the task of separating classes in feature space: w T x + b > 0 w T x + b = 0 w T x + b < 0 Department

More information

Support Vector Machines

Support Vector Machines Support Vector Machines Stephan Dreiseitl University of Applied Sciences Upper Austria at Hagenberg Harvard-MIT Division of Health Sciences and Technology HST.951J: Medical Decision Support Overview Motivation

More information

Geotechnical Earthquake Engineering

Geotechnical Earthquake Engineering Geotechnical Earthquake Engineering by Dr. Deepankar Choudhury Humboldt Fellow, JSPS Fellow, BOYSCAST Fellow Professor Department of Civil Engineering IIT Bombay, Powai, Mumbai 400 076, India. Email: dc@civil.iitb.ac.in

More information

Updated NGA-West2 Ground Motion Prediction Equations for Active Tectonic Regions Worldwide

Updated NGA-West2 Ground Motion Prediction Equations for Active Tectonic Regions Worldwide Updated NGA-West2 Ground Motion Prediction Equations for Active Tectonic Regions Worldwide Kenneth W. Campbell 1 and Yousef Bozorgnia 2 1. Corresponding Author. Vice President, EQECAT, Inc., 1130 NW 161st

More information

These slides follow closely the (English) course textbook Pattern Recognition and Machine Learning by Christopher Bishop

These slides follow closely the (English) course textbook Pattern Recognition and Machine Learning by Christopher Bishop Music and Machine Learning (IFT68 Winter 8) Prof. Douglas Eck, Université de Montréal These slides follow closely the (English) course textbook Pattern Recognition and Machine Learning by Christopher Bishop

More information

PACIFIC EARTHQUAKE ENGINEERING RESEARCH CENTER

PACIFIC EARTHQUAKE ENGINEERING RESEARCH CENTER PACIFIC EARTHQUAKE ENGINEERING RESEARCH CENTER Identification of Site Parameters that Improve Predictions of Site Amplification Ellen M. Rathje Sara Navidi Department of Civil, Architectural, and Environmental

More information

Choosing the Summary Statistics and the Acceptance Rate in Approximate Bayesian Computation

Choosing the Summary Statistics and the Acceptance Rate in Approximate Bayesian Computation Choosing the Summary Statistics and the Acceptance Rate in Approximate Bayesian Computation COMPSTAT 2010 Revised version; August 13, 2010 Michael G.B. Blum 1 Laboratoire TIMC-IMAG, CNRS, UJF Grenoble

More information

Estimation of Inelastic Response Spectra Using Artificial Neural Networks

Estimation of Inelastic Response Spectra Using Artificial Neural Networks Estimation of Inelastic Response Spectra Using Artificial Neural Networks J. Bojórquez & S.E. Ruiz Universidad Nacional Autónoma de México, México E. Bojórquez Universidad Autónoma de Sinaloa, México SUMMARY:

More information

CHARACTERIZATION OF DIRECTIVITY EFFECTS OBSERVED DURING 1999 CHI-CHI, TAIWAN EARTHQUAKE

CHARACTERIZATION OF DIRECTIVITY EFFECTS OBSERVED DURING 1999 CHI-CHI, TAIWAN EARTHQUAKE th World Conference on Earthquake Engineering Vancouver, B.C., Canada August -6, 4 Paper No. 74 CHARACTERIZATION OF DIRECTIVITY EFFECTS OBSERVED DURING 999 CHI-CHI, TAIWAN EARTHQUAKE Vietanh PHUNG, Gail

More information

Consortium of Organizations for Strong-Motion Observation Systems. COSMOS Annual Meeting Technical Session Summary

Consortium of Organizations for Strong-Motion Observation Systems. COSMOS Annual Meeting Technical Session Summary Consortium of Organizations for Strong-Motion Observation Systems COSMOS Annual Meeting Technical Session 2009 Summary Practical Application of the New ASCE 7-0 Required Procedures for Determining Site

More information

Hierarchical sparse Bayesian learning for structural health monitoring. with incomplete modal data

Hierarchical sparse Bayesian learning for structural health monitoring. with incomplete modal data Hierarchical sparse Bayesian learning for structural health monitoring with incomplete modal data Yong Huang and James L. Beck* Division of Engineering and Applied Science, California Institute of Technology,

More information

ANALYSIS OF THE CORRELATION BETWEEN INSTRUMENTAL INTENSITIES OF STRONG EARTHQUAKE GROUND MOTION

ANALYSIS OF THE CORRELATION BETWEEN INSTRUMENTAL INTENSITIES OF STRONG EARTHQUAKE GROUND MOTION ANALYSIS OF THE CORRELATION BETWEEN INSTRUMENTAL INTENSITIES OF STRONG EARTHQUAKE GROUND MOTION J.Enrique Martinez-Rueda 1, Evdokia Tsantali 1 1 Civil Engineering & Geology Division School of Environment

More information

CSC2541 Lecture 2 Bayesian Occam s Razor and Gaussian Processes

CSC2541 Lecture 2 Bayesian Occam s Razor and Gaussian Processes CSC2541 Lecture 2 Bayesian Occam s Razor and Gaussian Processes Roger Grosse Roger Grosse CSC2541 Lecture 2 Bayesian Occam s Razor and Gaussian Processes 1 / 55 Adminis-Trivia Did everyone get my e-mail

More information

Parametric Techniques Lecture 3

Parametric Techniques Lecture 3 Parametric Techniques Lecture 3 Jason Corso SUNY at Buffalo 22 January 2009 J. Corso (SUNY at Buffalo) Parametric Techniques Lecture 3 22 January 2009 1 / 39 Introduction In Lecture 2, we learned how to

More information

Relation of Pulse Period with Near-Fault Strong Motion Parameters

Relation of Pulse Period with Near-Fault Strong Motion Parameters th International Conference on Earthquake Geotechnical Engineering 1- November 15 Christchurch, New Zealand Relation of Pulse Period with Near-Fault Strong Motion Parameters V. Kardoutsou 1, P. Mimoglou,

More information

PATTERN RECOGNITION AND MACHINE LEARNING

PATTERN RECOGNITION AND MACHINE LEARNING PATTERN RECOGNITION AND MACHINE LEARNING Chapter 1. Introduction Shuai Huang April 21, 2014 Outline 1 What is Machine Learning? 2 Curve Fitting 3 Probability Theory 4 Model Selection 5 The curse of dimensionality

More information

Bayesian Linear Regression. Sargur Srihari

Bayesian Linear Regression. Sargur Srihari Bayesian Linear Regression Sargur srihari@cedar.buffalo.edu Topics in Bayesian Regression Recall Max Likelihood Linear Regression Parameter Distribution Predictive Distribution Equivalent Kernel 2 Linear

More information

STA 4273H: Sta-s-cal Machine Learning

STA 4273H: Sta-s-cal Machine Learning STA 4273H: Sta-s-cal Machine Learning Russ Salakhutdinov Department of Computer Science! Department of Statistical Sciences! rsalakhu@cs.toronto.edu! h0p://www.cs.utoronto.ca/~rsalakhu/ Lecture 2 In our

More information

DEPARTMENT OF COMPUTER SCIENCE Autumn Semester MACHINE LEARNING AND ADAPTIVE INTELLIGENCE

DEPARTMENT OF COMPUTER SCIENCE Autumn Semester MACHINE LEARNING AND ADAPTIVE INTELLIGENCE Data Provided: None DEPARTMENT OF COMPUTER SCIENCE Autumn Semester 203 204 MACHINE LEARNING AND ADAPTIVE INTELLIGENCE 2 hours Answer THREE of the four questions. All questions carry equal weight. Figures

More information

Gaussian processes. Chuong B. Do (updated by Honglak Lee) November 22, 2008

Gaussian processes. Chuong B. Do (updated by Honglak Lee) November 22, 2008 Gaussian processes Chuong B Do (updated by Honglak Lee) November 22, 2008 Many of the classical machine learning algorithms that we talked about during the first half of this course fit the following pattern:

More information

PILCO: A Model-Based and Data-Efficient Approach to Policy Search

PILCO: A Model-Based and Data-Efficient Approach to Policy Search PILCO: A Model-Based and Data-Efficient Approach to Policy Search (M.P. Deisenroth and C.E. Rasmussen) CSC2541 November 4, 2016 PILCO Graphical Model PILCO Probabilistic Inference for Learning COntrol

More information

Linear Regression. CSL603 - Fall 2017 Narayanan C Krishnan

Linear Regression. CSL603 - Fall 2017 Narayanan C Krishnan Linear Regression CSL603 - Fall 2017 Narayanan C Krishnan ckn@iitrpr.ac.in Outline Univariate regression Multivariate regression Probabilistic view of regression Loss functions Bias-Variance analysis Regularization

More information

Linear Regression. CSL465/603 - Fall 2016 Narayanan C Krishnan

Linear Regression. CSL465/603 - Fall 2016 Narayanan C Krishnan Linear Regression CSL465/603 - Fall 2016 Narayanan C Krishnan ckn@iitrpr.ac.in Outline Univariate regression Multivariate regression Probabilistic view of regression Loss functions Bias-Variance analysis

More information

Stat 5101 Lecture Notes

Stat 5101 Lecture Notes Stat 5101 Lecture Notes Charles J. Geyer Copyright 1998, 1999, 2000, 2001 by Charles J. Geyer May 7, 2001 ii Stat 5101 (Geyer) Course Notes Contents 1 Random Variables and Change of Variables 1 1.1 Random

More information

Reliability of Acceptance Criteria in Nonlinear Response History Analysis of Tall Buildings

Reliability of Acceptance Criteria in Nonlinear Response History Analysis of Tall Buildings Reliability of Acceptance Criteria in Nonlinear Response History Analysis of Tall Buildings M.M. Talaat, PhD, PE Senior Staff - Simpson Gumpertz & Heger Inc Adjunct Assistant Professor - Cairo University

More information

Parametric Techniques

Parametric Techniques Parametric Techniques Jason J. Corso SUNY at Buffalo J. Corso (SUNY at Buffalo) Parametric Techniques 1 / 39 Introduction When covering Bayesian Decision Theory, we assumed the full probabilistic structure

More information

Usability of the Next Generation Attenuation Equations for Seismic Hazard Assessment in Malaysia

Usability of the Next Generation Attenuation Equations for Seismic Hazard Assessment in Malaysia Azlan Adnan, Patrick Liq Yee Tiong, Yue Eng Chow/ International Journal of Engineering Vol. 2 Issue 1, Jan-Feb 212, pp.639-644 Usability of the Next Generation Attenuation Equations for Seismic Hazard

More information

Lecture 5: GPs and Streaming regression

Lecture 5: GPs and Streaming regression Lecture 5: GPs and Streaming regression Gaussian Processes Information gain Confidence intervals COMP-652 and ECSE-608, Lecture 5 - September 19, 2017 1 Recall: Non-parametric regression Input space X

More information

Linear Regression and Its Applications

Linear Regression and Its Applications Linear Regression and Its Applications Predrag Radivojac October 13, 2014 Given a data set D = {(x i, y i )} n the objective is to learn the relationship between features and the target. We usually start

More information

GROUND MOTION TIME HISTORIES FOR THE VAN NUYS BUILDING

GROUND MOTION TIME HISTORIES FOR THE VAN NUYS BUILDING GROUND MOTION TIME HISTORIES FOR THE VAN NUYS BUILDING Prepared for the PEER Methodology Testbeds Project by Paul Somerville and Nancy Collins URS Corporation, Pasadena, CA March 7, Site Conditions The

More information