Bayesian Modeling of log N log S
|
|
- Lorraine McDonald
- 5 years ago
- Views:
Transcription
1 Bayesian Modeling of log N log S Irina S Udaltsova Vasileios Stampoulis Department of Statistics University of California, Davis July 26th, / 44
2 Introduction Scientific Objectives: Develop a comprehensive method to infer (properties of) the distribution of source fluxes for a wide variety source populations. Statistical Objectives: Inference: Account for non-ignorable missing data Model Checking: Evaluate the adequacy of a given model Model Selection: Select the best model for a given dataset Collaborators: Paul D. Baines (UCD), Andreas Zezas (University of Crete & CfA), Vinay Kashyap (CfA). 2 / 44
3 CHANDRA:
4 Introduction: log N log S Cumulative number of sources detectable at a given sensitivity: N(> S) = i I {Si >S} or the number of source fluxes brighter than a threshold, S. The name log N log S refers to the relationship between (or plot of) log 10 N(> S) and log 10 S. In many cosmological applications there is strong theory that expects the log N log S to obey a Power law: N(> S) = N I {Si >S} αs θ, S > S min i=1 Taking the logarithm gives the linear log(n) log(s) relationship. 4 / 44
5 The Rationale for log N log S Fitting Knowing the specific relationship for different objects (e.g., stars, galaxies, pulsars) gives a lot of information about the underlying physics (e.g., the mass of galaxies). Helps in tests for cosmological parameters, constrains evolutionary models, etc. Primary Goal: Estimate θ, the power law slope, while properly accounting for detector uncertainties and biases. Note: There is uncertainty on both x- and y-axes (i.e., N and S). 5 / 44
6 Flux Distribution Probabilistic Connection: Under independent sampling, linearity on the log N log S scale is equivalent to the flux distribution being a Pareto distribution. S i τ 1, θ iid Pareto (θ, τ1 ), i = 1,..., N. Piecewise-linear power-law extends to broken power-law model for flux distribution, subject to continuity constraint at breakpoint τ 2 : { α0 θ log 10 (1 F G (s)) = 0 log 10 (s) τ 1 < s τ 2 α 1 θ 1 log 10 (s) s > τ 2 ( [ ( ) ]) θ1 τ1 Y I X 1 + (1 I ) X 2, where: I Bin 1, 1 τ 2 X 1 Truncated-Pareto (τ 1, θ 1, τ 2 ), X 2 Pareto (τ 2, θ 2 ). Can be extended to a multiple broken power-law with arbitrary number, m 1, of break-points. 6 / 44
7 Missing Data There are many potential causes of uncertainty and missing data in astronomical data: Low-count sources (below detection threshold) Detector schedules (source not within detector range) Background contamination (e.g., total=source+background) Foreground contamination (other objects between the source and detector) etc. Important: Whether a source is observed is a function of its source count (intensity), which is unobserved for unobserved sources missing data mechanism is non-ignorable, so it needs to be carefully accounted for in the analysis. 7 / 44
8 The Data Flux is not measured directly; instead, the data are a list of photon counts, with some extra information about the background and detector properties. Src_ID Counts Bgr_intensity Src_area Off_axis_L Effective_area and an incompleteness function (selection function), specifying the probability of source detection under a range of conditions: P( Detecting a source with flux S, background intensity B, location L and effective area E ) g(s, B, L, E) 8 / 44
9 Single Power-Law Model Standard power-law flux distribution: S i τ, θ iid Pareto (θ, τ), i = 1,..., N. Source and background photon counts: Yi tot indep S i, B i, L i, E i Pois (λ(s i, B i, L i, E i ) + k(b i, L i, E i )), i = 1,..., N. Incompleteness, missing data indicators: Prior distributions: I i Bernoulli (g (S i, B i, L i, E i )). N Neg-Bin (a N, b N ), θ Gamma(a, b), τ Gamma(a m, b m ). 9 / 44
10 Broken Power-Law Model Broken power-law flux distribution: S i τ, θ iid Broken-Pareto Source and background photon counts: ( θ, τ ), i = 1,..., N. Yi tot indep S i, B i, L i, E i Pois (λ(s i, B i, L i, E i ) + k(b i, L i, E i )), i = 1,..., N. Incompleteness, missing data indicators: Prior distributions: τ j = τ 1 + I i Bernoulli (g (S i, B i, L i, E i )). N Neg-Bin (a N, b N ), θ j indep Gamma(a j, b j ), j = 1,..., m, j k=2 τ 1 Gamma(a m, b m ) e η k, η j indep Normal(µ j, c j ), j = 2,..., m. 10 / 44
11 Model Overview Unusual points and important notes: The dimension of the missing data is unknown (care must be taken with conditioning) The incompleteness function g must be known well; can take any form; is problem-specific The number of flux populations in sample, m, must be specified The flux lower limit and break-points, τ j, j = 1,..., m, can be estimated Prior parameters can be science-based, i.e., weakly informative 11 / 44
12 Posterior Inference (Single Pareto) Inference about θ, N, S, τ is based on the observed data posterior distribution. Care must be taken with the variable dimension marginalization over the unobserved fluxes. The single power-law posterior can be shown to be: p ( N, θ, τ, S obs, Yobs n, src Yobs tot ), B obs, L obs, E obs ( N n ) ( I {n N} (1 π(θ, τ)) (N n) N + an 1 a N 1 ) ( b N b a Γ(a) θa 1 e bθ b am n m I {θ>0} Γ(a τ am 1 e bmτ I {τ>0} m) i=1 g(s i, B i, L i, E i ) (λ i + k i ) Y i tot ( Y tot i Y src i ) ( λ i λ i + k i Y i tot! ) Y src ( i e (λ i +k i ) I {Y tot Z i + } 1 λ i λ i + k i ) Y tot i Y i src I { Y src i ) N ( bn 1 + b N ) an I {N Z + } p (B i, L i, E i ) θτ θ S (θ+1) I i {τ<si } {0,1,...,Y i tot } } with λ i λ(s i, B i, L i, E i ) and k i k(b i, L i, E i ). 12 / 44
13 Posterior Inference (Broken-Pareto) The broken power-law posterior can be shown to be: p ( N, θ, τ, S obs, Yobs n, src Yobs tot ), B obs, L obs, E obs [( ) ] 1 N = p(n, Y obs tot, B obs, L obs, E obs ) I n {n N} (1 π(θ, τ)) (N n) ) ( ( ) N ( ) an N + an 1 1 bn m b a j I a N 1 {N Z 1 + b N 1 + b + } j N Γ(a j=1 j ) θa j 1 e b j θj I j {θj >0} n p (τ 1,..., τ m) I {0<τ1 <τ 1 < <τm} p (B i, L i, E i ) g(s i, B i, L i, E i ) i=1 m j 1 ( ) θl ( ) ( ) τl+1 (θj +1) θj Si I τ j=1 l=1 l { } τ j τ τ j j S i <τ j+1 (λ i + k i ) Y i tot Y i tot! ( ) ( Y tot λ i i Yi src λ i + k i ) Y src i e (λ i +k i ) I {Y tot Z i + } ( 1 λ i λ i + k i ) Y tot i Y i src I { Y src i {0,1,...,Y i tot } } with τ m+1 = +, λ i λ(s i, B i, L i, E i ), k i k(b i, L i, E i ), and 0 i=1 ( τi+1 τ i ) θi = / 44
14 Computation Strategy: Gibbs Sampler The Gibbs sampler consists of five steps: [N n, θ], [θ n, N, S obs, τ], [τ n, N, θ, S obs, B obs, L obs, E obs ], [ Sobs N, θ, τ, I obs, Yobs tot, Y obs src, B ] obs, L obs, E obs, [ Y src tot obs Yobs, B ] obs, L obs, E obs, I obs, S obs where each parameter block is sampled via various numerical and probabilistic routines, including Metropolis Hastings. To Model Checking To Model Validation To Computational Details 14 / 44
15 Simulation Example MCMC Output (L) Posterior logn-logs (red: missing, gray: observed), truth (blue). (R) Posterior distributions for N, θ, τ 15 / 44
16 Non-Ignorable Missingness (L) Nonignorable (full) analysis: (R) Ignorable analysis: ˆθ = 0.990, (0.803, 1.192) ˆθ = 0.784, (0.520, 0.978) Truth: θ = / 44
17 Posterior Predictive Checking Consider testing the hypothesis: H 0 : The model is correctly specified, vs., H 1 : The model is not correctly specified. MCMC draws allow to check the adequacy of the model fit for Bayesian models using the posterior predictive distribution: p(y y) = p(y, θ y)dθ = p(y θ) p(θ y)dθ Idea: (Assuming conditional independence) We expect the predictive distribution of new data to look similar to the empirical distribution of the observed data. Extension: We expect function summaries of interesting features (e.g., test statistics) of the new data to look similar to the empirical distribution of functions of the observed data. 17 / 44
18 Posterior Predictive Checking Procedure: 1. Sample θ from the posterior distribution p(θ y) 2. Given θ, sample y (replicate photon counts) from p(y θ) 3. Compute desired statistic T (y ) 4. Compute the posterior predictive p value = proportion of the replicate cases in which T (y ) exceeds T (y): p b = P (T (y ) T (y) y, H 0 ). Choice of test statistic T (y) to summarize the desired features of distribution of the photon counts? n = dim(y 1,..., y n ) Max = max(y 1,..., y n ) Med = Q 2 (y 1,..., y n ) IQR = Q 3 (y 1,..., y n ) Q 1 (y 1,..., y n ) Skewness = skewness(y 1,..., y n ) R 2 S crude = coefficient of determination between log N log S crude, where S crude = (y bg)γ/e 18 / 44
19 Posterior Predictive Check: Examples If the model is true, the posterior predictive p-value will almost certainly be very close to 0.5. Figure: (Left) If the model is in fact true, the sample IQR of new sets of n observations will look similar to that of observed n data points. (Right) The sample value of RS 2 crude of new sets of n observations does not look similar to that of observed n data points, so the data is unlikely to have been produced from current model.
20 Bivariate PP p values The bivariate posterior predictive p value = area under slice of bivariate posterior predictive distr. of two test statistics Figure: (Left) Model correctly specified, (Right) Model incorrectly specified 20 / 44
21 Model Selection Figure: (Left) Regular Pareto, (Right) Broken Pareto Note: similar posterior log N log S, but very different predictive and inferential properties 21 / 44
22 Model Selection Model selection is a challenging problem in Bayesian applications. Popular methods are: Bayes Factor, DIC, and WAIC. Bayes Factor p(m 1 y) p(m 2 y) = p(y M 1) p(m 1 ) p(y M 2 ) p(m 2 ) = BF p(m 1 ) 12 p(m 2 ), Has good model selection performance in simple settings Problem: Requires evaluating marginal probability of the data given the model M k p(y M k ) = p(y) = p(y θ)p(θ)dθ. Ω 22 / 44
23 Model Selection DIC, Deviance Information Criteria (Spiegelhalter, 2002) DIC = 2 log p(y θ Bayes ) + 2p DIC { } p DIC = 2 log p(y θ Bayes ) E[log p(y θ) y] WAIC, Widely Applicable Information Criteria (Watanabe, 2010) N WAIC = 2 log E[p(y i θ) y] + 2p WAIC p WAIC = 2 i=1 N {log E[p(y i θ) y] E[log p(y i θ) y]} i=1 Problem: In complex hierarchical problems, DIC and WAIC can have poor model selection performance by over-fitting. Alternatives? 23 / 44
24 Adaptive Fence Method (AF) Original method of adaptive fence (Jiang et.al., 2008). For a collection of nested candidate models M 1, M 2,..., M k, with the full model M k, Select a measure of goodness of fit (Q = log p(y ˆθ MLE )) Select optimality criterion (minimal dimension) Construct a fence Q(M j ) Q( M) c where c is a constant, and necessarily Q(M j ) > Q( M) for all j Label model M c if it is in the Fence and satisfied optimality criterion The model to maximize p = P(M c = M opt ) is chosen to be the correct model The probability P is evaluated under model M 24 / 44
25 Adaptive Fence Method (AF) Recent article (Jiang, 2014) presents various fence methods. Example of Adaptive Fence Method p.star[, kk] c.vec[, kk] Adaptive Fence (AF) selects the cut-off c by maximizing the empirical probability of selection Parametric bootstrap is used to evaluate the maximum empirical probability P given parameter MLE of model M Problem: MLE are not easily available for log(n) log(s) application with missing data Problem: AF has adequate performance when true model is (closest to) the boundary model of the nested candidate models Problem: Non-nested candidate models, for which no full model can be easily identified? 25 / 44
26 Bayesian Adaptive Fence Method (BAFM) We extend AF method for Bayesian applications and propose Use non-parametric bootstrap Use alternate (Bayesian version) measures of goodness-of-fit, Q 1 = N log i=1 [ 1 L L l=1 Q 2 = log p(y β PostMean ), Q 3 = log p(y β PostMedian ), p(y i β (l) MCMC ) Q 4 = log p(y β PostMode ), Q 5 = 1 L log p(y β (l) MCMC L ), l=1 { } Q 6 = Median log p(y β (l) MCMC ), l = 1,..., L. where l = 1,..., L draws of MCMC parameters θ (l) are collected. Introduce two phantom boundary models of best and worst fit Q( M k+1 ) = 0 and Q(M 0 ) > max{q(m 1 ),..., Q(M k )}. 26 / 44 ],
27 Example: p vs c plot (BAFM) p* based on Q6. N = 200. p* c 27 / 44
28 Simulation of Model Selection Performance Simulated true model SimID Truth τ 1, τ 2, τ 3 θ 1, θ 2, θ 3 1 bp ,, 0.5,, 2 bp , , 0.5, 0.7, 3 bp , , 0.5, 1.5, 4 bp , , , 1.0, bp , , , 0.7, 0.9 Results SimID Truth DIC mean DIC median DIC mode BF BAFM 1 bp (Q 4 ) 2 bp (Q 4 ) 3 bp (Q 4 ) 4 bp (Q 4 ) 5 bp (Q 2 ) Table: Proportion of correctly selecting true model based on DIC at posterior mean, median, mode, BF with harmonic mean approximation, and Bayesian Adaptive Fence Method for best performing measure Q with automatic peak detection. 28 / 44
29 BAFM Issues Which criterion measure Q to use? Complete likelihood is unknown Middle peak does not always exist. Cannot select one best model uniquely Very computationally intensive. Need to run MCMC for each bootstrap re-sample 29 / 44
30 References P.D. Baines, I.S. Udaltsova, A. Zezas, V.L. Kashyap (2011) Bayesian Estimation of log N log S, Proc. of Statistical Challenges in Modern Astronomy V P.D. Baines, I.S. Udaltsova, A. Zezas, V.L. Kashyap (2016) Hierarchical Modeling for log N log S: Theory, Methods and Applications Working Paper J. Jiang. et. al (2008) Fence methods for mied model selection. Ann Stat, 36: J. Jiang (2014) The Fence Methods. Adv Stat, 2014: R.J.A. Little, D.B. Rubin. (2002) Statistical analysis with missing data, Wiley. R.K.W. Wong, P. Baines, A. Aue, T.C.M. Lee, V.L. Kashyap (2014) Automatic estimation of flux distributions of astrophysical source populations. Ann Appl Stat, 8:1690?1712. A. Zezas et al. (2004) Chandra survey of the Bar region of the SMC Revista Mexicana de Astronomia y Astrofisica (Serie de Conferencias) Vol. 20. IAU Colloquium 194, pp / 44
31 Validating Bayesian Computation Given the complexity of the hierarchical model and computation, it is important to validate that everything works correctly: a self-consistency check. Procedure: 1. Simulate parameters from the prior, and data from the model, given those parameters 2. Fit the model to obtain posterior intervals 3. Record whether or not the true value of the parameter was within the interval 4. Repeat steps 1, 2, & 3 a large number of times, and calculate the average coverage The average and nominal coverages should be equal. These validation checks are extremely important when dealing with complex procedures. 31 / 44
32 Model Validation 90% Posterior intervals vs. Truth (200 datasets) Actual vs. Nominal Coverage (200 datasets) Posterior 0e+00 1e 17 2e 17 3e 17 4e 17 5e flagged intervals Actual Target N theta Smin Sobs 0e+00 1e 17 2e 17 3e 17 4e 17 5e True tau.1 Nominal Figure: (Left) Posterior 90% intervals of τ 1 = S min show reasonable coverage roughly 90% of the time, when nominal coverage is set to 90%. (Right) Average coverage probabilities for each parameter is within the expected bounds at each nominal probability. 32 / 44
33 Computational Details The Gibbs sampler consists of five steps: [N n, θ], [θ n, N, S obs, τ], [τ n, N, θ, S obs, B obs, L obs, E obs ], [ Sobs N, θ, τ, I obs, Yobs tot, Yobs, src ] B obs, L obs, E obs, [ Y src obs Yobs tot ], B obs, L obs, E obs, I obs, S obs. Sample the total number of sources, N, (Numerical Integration): ( ) N p (N ) I n {n N} (1 π(θ, τ)) (N n) p (N) p (S obs N, θ, τ) Γ(N + a ( ) N N) 1 Γ(N n + 1) (1 π(θ, τ)) (N n) I {n N} b N + 1 The marginal probability of observing a source π(θ, τ) is pre-computed via the numerical integration. π(θ, τ) = g(s i, B i, L i, E i ) p(s i θ, τ) p(b i, L i, E i ) ds i db i dl i de i. 33 / 44
34 Computational Details cont... The Gibbs sampler consists of five steps: [N n, θ], [θ n, N, S obs, τ], [τ n, N, θ, S obs, B obs, L obs, E obs ], [ Sobs N, θ, τ, I obs, Yobs tot, Yobs, src ] B obs, L obs, E obs, [ Y src obs Yobs tot ], B obs, L obs, E obs, I obs, S obs. Sample the power-law slope, θ, (Metropolis Hastings using a Normal-proposal): p (θ ) p (θ) p (S obs N, θ, τ) (1 π(θ, τ)) (N n) ( n ( ) ) (1 π(θ, τ)) (N n) Si Gamma θ; a + n, b + log τ Sample the observed photon counts Yobs,i src, i = 1,..., n: p (Y src i ) p(yi src ( Bin Y tot i, S i, B i, L i, E i ) Yi src ; Yi tot, i=1 λ(s i, B i, L i, E i ) λ(s i, B i, L i, E i ) + k(b i, L i, E i ) ) 34 / 44
35 Computational Details cont... Sample the minimum flux τ (Metropolis Hastings using a Truncated-Normal-proposal after log-transformation): p(τ ) p(τ) p(θ) p(n) p (B obs, L obs, E obs ) p (n, S obs, I obs N, θ, τ, B obs, L obs, E obs ) Gamma (τ; a m + nθ, b m ) (1 π(θ, τ)) N n I {0<τ<cm} p(η = log(τ) ) = e η p(τ = e η ) e η(nθ+am+1) e bmeη (1 π(θ, τ = e η )) N n I {η<log(cm)}. Sample the fluxes S obs,i, i = 1,..., n (Metropolis Hastings using a Normal-proposal): p (S i ) p (S i N, θ, τ) p (I i = 1 S i, B i, L i, E i ) p ( Yi tot S i, B i, L i, E i ) p ( Yi src Yi tot ), S i, B i, L i, E i Pareto(S i ; θ, τ) g(s i, B i, L i, E i ) Pois(Yi tot ; λ(s i, B i, L i, E i )+k(b i, L i, E i )) ( Bin Yi src ; Yi tot, λ(s i, B i, L i, E i ) λ(s i, B i, L i, E i ) + k(b i, L i, E i ) ) 35 / 44
36 Computational Details cont... Sample θ = (θ1,..., θ m) T : (Metropolis-Hastings using a Normal-proposal) [ p(θ ) (1 π(θ, τ)) N n] ( ) m τj+1 m ] Gamma θ j ; a j + n(j) 1, b j + I {j m} log [n(i)i {i j+1} + ( ) si log, τ j=1 j τ i=1 i I(j) j where I(j) = { i : τ j s i < τ j+1 } and n(j) is the cardinality of I(j) i.e., I(j) (n(j)) denotes the set (number) of source indices whose flux is contained in the interval corresponding to the j-th mixture component. Sample the break-points τ = (τ2,..., τ m) T (Metropolis-Hastings based on original transformed scale η j = h( τ τ 1 ) = log(τ j τ j 1 ), j = 2,..., m): [ p(η ) = p(h( τ τ 1 ) ) (1 π(θ, τ)) (N n)] exp 1 m { cj (η j µ j ) } 2 I {τ1 <τ 2 2 < <τm} j=2 m j 1 ( ) θl τl+1 n(j) ( ) ( ) (θj +1) θj si I τ j=1 l=1 l {τ1 <min(s τ i I(j) j τ 1,...,sn)}, j 36 / 44
37 Computational Details cont... Fluxes of missing sources can (optionally) be imputed to produce posterior draws of a corrected log N log S Impute missing fluxes S mis,i, i = 1,..., n (Rejection Sampling): (B i, L i, E i ) p(b i, L i, E i ) S i n, N, θ, τ, B i, L i, E i, I i = 0 (1 g(s i, B i, L i, E i )) Pareto(S i ; θ, τ). 37 / 44
38 Computational Notes Some important things to note: For single power-law models computation is relatively fast (few hours), and insensitive to the number of missing sources The fluxes of the missing sources need not be imputed Fluxes of missing sources can (optionally) be imputed to produce posterior draws of a corrected log N log S Computation for the broken-power law model is slower (day) Generalized mixtures of Pareto s (or other forms) require only minor modifications of general scheme 38 / 44
39 Application: CDFN X-rays Application: X-ray emission from point sources in the CHANDRA Deep Field North (CDFN) survey. The deepest kev band survey ever made 2 Ms exposure covering field of 448 sq. arcmin Subsample sources based on off-axis angle threshold of 8 arcmin Dataset consists of 225 observed sources Priors: a = 10, b = 10, E [N] = 300, Var(N) = 100 2, E [τ 1 ] = , Var(τ 1 ) = ( ) 2, E [log(τ 2 τ 1 )] = 38, Var(log(τ 2 τ 1 )) = / 44
40 CDFN log(n)-log(s)
41 Method bp0 bp1 bp2 bp0 g = 1 BAFM Sel. DIC.Mean Q 1 bp2 DIC.Median Q 2 bp2 DIC.Mode Q 3 bp2 DIC.V Q 4 bp2 WAIC Q 5 bp2 WAIC Q 6 bp2 Table: DIC, WAIC, BAFM for CDFN analysis.
42 bp0 bp1 bp2 g = 1 # obs src (0.63) 0.07 (0.08) 0.23 Minimum ph ct (0.40) 0.40 (0.41) 0.36 Maximum ph ct (0.26) 0.22 (.24) 0.08 Median ph ct (0.10) 0.03 (0.05) 0.40 Lower quartile of ph cts (0.23) 0.17 (0.17) 0.28 Upper quantile of ph cts (0.12) 0.04 (0.07) 0.32 IQR of ph cts (0.11) 0.04 (0.07) 0.32 Crude estimate of R (0.33) 0.31 (0.31) 0.15 # obs src vs. Median ph ct (0.22) 0.15 (0.20) 0.65 Lower vs. Upper quartile (0.26) 0.16 (0.21) 0.75 # obs sources vs. IQR (0.20) 0.13 (0.20) 0.72 # obs sources vs. R (0.59) 0.59 (0.60) 0.18 Table: Posterior predictive p-values for CDFN analysis. (Adjusted for low-flux*)
43 CDFN Conclusions Estimate 95% (central) (Wong et.al.,2014) Parameter Mean (Med) interval Estimate SE θ (0.59) (0.40,1.09) θ (0.56) (0.36,1.18) θ (0.82) (0.67,1.03) log 10 (τ 1 ) (-16.44) (-16.60,-16.35) log 10 (τ 2 ) (-16.11) (-16.36,-15.61) log 10 (τ 3 ) (-15.74) (-16.00,-15.47) N 281 (280) (259,312) Table: Parameter estimates for CDFN analysis based on bp2 model. Compared to competing method for estimation of the number of breakpoints and parameters via interwoven EM (Wong et.al., 2014) 43 / 44
44 CDFN Conclusions Estimated slope of 0.46 (median), (mean) 95% (central) interval = (0.312, 0.681) Wong et al. (2014) estimated broken power-law slopes of ˆθ = (0.483, 0.854). Estimated missing data fraction 24%, interval:(0.0, 28.25%) Evidence of a possible break in the power-law in the observed log N log S. Note: The ignorable analysis gives a posterior median of 0.363, mean of 0.367, and 95% interval (0.248, 0.511). 44 / 44
Bayesian Estimation of log N log S
Bayesian Estimation of log N log S Paul Baines, Irina Udaltsova Department of Statistics University of California, Davis July 12th, 2011 Introduction What is log N log S? Cumulative number of sources detectable
More informationBayesian Estimation of log N log S
Bayesian Estimation of log N log S Paul D. Baines Department of Statistics University of California, Davis May 10th, 2013 Introduction Project Goals Develop a comprehensive method to infer (properties
More informationAges of stellar populations from color-magnitude diagrams. Paul Baines. September 30, 2008
Ages of stellar populations from color-magnitude diagrams Paul Baines Department of Statistics Harvard University September 30, 2008 Context & Example Welcome! Today we will look at using hierarchical
More informationST440/540: Applied Bayesian Statistics. (9) Model selection and goodness-of-fit checks
(9) Model selection and goodness-of-fit checks Objectives In this module we will study methods for model comparisons and checking for model adequacy For model comparisons there are a finite number of candidate
More informationLuminosity Functions
Department of Statistics, Harvard University May 15th, 2012 Introduction: What is a Luminosity Function? Figure: A galaxy cluster. Introduction: Project Goal The Luminosity Function specifies the relative
More informationFitting Narrow Emission Lines in X-ray Spectra
Outline Fitting Narrow Emission Lines in X-ray Spectra Taeyoung Park Department of Statistics, University of Pittsburgh October 11, 2007 Outline of Presentation Outline This talk has three components:
More informationPrinciples of Bayesian Inference
Principles of Bayesian Inference Sudipto Banerjee University of Minnesota July 20th, 2008 1 Bayesian Principles Classical statistics: model parameters are fixed and unknown. A Bayesian thinks of parameters
More informationBayesian Computation in Color-Magnitude Diagrams
Bayesian Computation in Color-Magnitude Diagrams SA, AA, PT, EE, MCMC and ASIS in CMDs Paul Baines Department of Statistics Harvard University October 19, 2009 Overview Motivation and Introduction Modelling
More informationIntroduction to Probabilistic Machine Learning
Introduction to Probabilistic Machine Learning Piyush Rai Dept. of CSE, IIT Kanpur (Mini-course 1) Nov 03, 2015 Piyush Rai (IIT Kanpur) Introduction to Probabilistic Machine Learning 1 Machine Learning
More informationMeasurement Error and Linear Regression of Astronomical Data. Brandon Kelly Penn State Summer School in Astrostatistics, June 2007
Measurement Error and Linear Regression of Astronomical Data Brandon Kelly Penn State Summer School in Astrostatistics, June 2007 Classical Regression Model Collect n data points, denote i th pair as (η
More informationpyblocxs Bayesian Low-Counts X-ray Spectral Analysis in Sherpa
pyblocxs Bayesian Low-Counts X-ray Spectral Analysis in Sherpa Aneta Siemiginowska Harvard-Smithsonian Center for Astrophysics Vinay Kashyap (CfA), David Van Dyk (UCI), Alanna Connors (Eureka), T.Park(Korea)+CHASC
More informationNew Bayesian methods for model comparison
Back to the future New Bayesian methods for model comparison Murray Aitkin murray.aitkin@unimelb.edu.au Department of Mathematics and Statistics The University of Melbourne Australia Bayesian Model Comparison
More informationOverlapping Astronomical Sources: Utilizing Spectral Information
Overlapping Astronomical Sources: Utilizing Spectral Information David Jones Advisor: Xiao-Li Meng Collaborators: Vinay Kashyap (CfA) and David van Dyk (Imperial College) CHASC Astrostatistics Group April
More informationAccounting for Calibration Uncertainty
: High Energy Astrophysics and the PCG Sampler 1 Vinay Kashyap 2 Taeyoung Park 3 Jin Xu 4 Imperial-California-Harvard AstroStatistics Collaboration 1 Statistics Section, Imperial College London 2 Smithsonian
More informationMetropolis-Hastings Algorithm
Strength of the Gibbs sampler Metropolis-Hastings Algorithm Easy algorithm to think about. Exploits the factorization properties of the joint probability distribution. No difficult choices to be made to
More informationApproximate Bayesian computation for spatial extremes via open-faced sandwich adjustment
Approximate Bayesian computation for spatial extremes via open-faced sandwich adjustment Ben Shaby SAMSI August 3, 2010 Ben Shaby (SAMSI) OFS adjustment August 3, 2010 1 / 29 Outline 1 Introduction 2 Spatial
More informationAccounting for Calibration Uncertainty in Spectral Analysis. David A. van Dyk
Accounting for in Spectral Analysis David A van Dyk Statistics Section, Imperial College London Joint work with Vinay Kashyap, Jin Xu, Alanna Connors, and Aneta Siegminowska IACHEC March 2013 David A van
More informationCSC321 Lecture 18: Learning Probabilistic Models
CSC321 Lecture 18: Learning Probabilistic Models Roger Grosse Roger Grosse CSC321 Lecture 18: Learning Probabilistic Models 1 / 25 Overview So far in this course: mainly supervised learning Language modeling
More informationSpatial Statistics Chapter 4 Basics of Bayesian Inference and Computation
Spatial Statistics Chapter 4 Basics of Bayesian Inference and Computation So far we have discussed types of spatial data, some basic modeling frameworks and exploratory techniques. We have not discussed
More informationBayesian Model Diagnostics and Checking
Earvin Balderama Quantitative Ecology Lab Department of Forestry and Environmental Resources North Carolina State University April 12, 2013 1 / 34 Introduction MCMCMC 2 / 34 Introduction MCMCMC Steps in
More informationBayesian Learning. HT2015: SC4 Statistical Data Mining and Machine Learning. Maximum Likelihood Principle. The Bayesian Learning Framework
HT5: SC4 Statistical Data Mining and Machine Learning Dino Sejdinovic Department of Statistics Oxford http://www.stats.ox.ac.uk/~sejdinov/sdmml.html Maximum Likelihood Principle A generative model for
More informationParameter Estimation. William H. Jefferys University of Texas at Austin Parameter Estimation 7/26/05 1
Parameter Estimation William H. Jefferys University of Texas at Austin bill@bayesrules.net Parameter Estimation 7/26/05 1 Elements of Inference Inference problems contain two indispensable elements: Data
More informationAccounting for Calibration Uncertainty in Spectral Analysis. David A. van Dyk
Bayesian Analysis of Accounting for in Spectral Analysis David A van Dyk Statistics Section, Imperial College London Joint work with Vinay Kashyap, Jin Xu, Alanna Connors, and Aneta Siegminowska ISIS May
More informationAccounting for Complex Sample Designs via Mixture Models
Accounting for Complex Sample Designs via Finite Normal Mixture Models 1 1 University of Michigan School of Public Health August 2009 Talk Outline 1 2 Accommodating Sampling Weights in Mixture Models 3
More informationStatistical Tools and Techniques for Solar Astronomers
Statistical Tools and Techniques for Solar Astronomers Alexander W Blocker Nathan Stein SolarStat 2012 Outline Outline 1 Introduction & Objectives 2 Statistical issues with astronomical data 3 Example:
More informationBayesian Methods for Machine Learning
Bayesian Methods for Machine Learning CS 584: Big Data Analytics Material adapted from Radford Neal s tutorial (http://ftp.cs.utoronto.ca/pub/radford/bayes-tut.pdf), Zoubin Ghahramni (http://hunch.net/~coms-4771/zoubin_ghahramani_bayesian_learning.pdf),
More informationLikelihood-based inference with missing data under missing-at-random
Likelihood-based inference with missing data under missing-at-random Jae-kwang Kim Joint work with Shu Yang Department of Statistics, Iowa State University May 4, 014 Outline 1. Introduction. Parametric
More informationFully Bayesian Analysis of Calibration Uncertainty In High Energy Spectral Analysis
In High Energy Spectral Analysis Department of Statistics, UCI February 26, 2013 Model Building Principle Component Analysis Three Inferencial Models Simulation Quasar Analysis Doubly-intractable Distribution
More informationBayesian non-parametric model to longitudinally predict churn
Bayesian non-parametric model to longitudinally predict churn Bruno Scarpa Università di Padova Conference of European Statistics Stakeholders Methodologists, Producers and Users of European Statistics
More informationBAYESIAN METHODS FOR VARIABLE SELECTION WITH APPLICATIONS TO HIGH-DIMENSIONAL DATA
BAYESIAN METHODS FOR VARIABLE SELECTION WITH APPLICATIONS TO HIGH-DIMENSIONAL DATA Intro: Course Outline and Brief Intro to Marina Vannucci Rice University, USA PASI-CIMAT 04/28-30/2010 Marina Vannucci
More informationFractional Imputation in Survey Sampling: A Comparative Review
Fractional Imputation in Survey Sampling: A Comparative Review Shu Yang Jae-Kwang Kim Iowa State University Joint Statistical Meetings, August 2015 Outline Introduction Fractional imputation Features Numerical
More informationEM Algorithm II. September 11, 2018
EM Algorithm II September 11, 2018 Review EM 1/27 (Y obs, Y mis ) f (y obs, y mis θ), we observe Y obs but not Y mis Complete-data log likelihood: l C (θ Y obs, Y mis ) = log { f (Y obs, Y mis θ) Observed-data
More informationBayesian Inference in Astronomy & Astrophysics A Short Course
Bayesian Inference in Astronomy & Astrophysics A Short Course Tom Loredo Dept. of Astronomy, Cornell University p.1/37 Five Lectures Overview of Bayesian Inference From Gaussians to Periodograms Learning
More informationTwo Statistical Problems in X-ray Astronomy
Two Statistical Problems in X-ray Astronomy Alexander W Blocker October 21, 2008 Outline 1 Introduction 2 Replacing stacking Problem Current method Model Further development 3 Time symmetry Problem Model
More informationBayesian search for other Earths
Bayesian search for other Earths Low-mass planets orbiting nearby M dwarfs Mikko Tuomi University of Hertfordshire, Centre for Astrophysics Research Email: mikko.tuomi@utu.fi Presentation, 19.4.2013 1
More informationMarkov Chain Monte Carlo methods
Markov Chain Monte Carlo methods By Oleg Makhnin 1 Introduction a b c M = d e f g h i 0 f(x)dx 1.1 Motivation 1.1.1 Just here Supresses numbering 1.1.2 After this 1.2 Literature 2 Method 2.1 New math As
More informationSTA 4273H: Statistical Machine Learning
STA 4273H: Statistical Machine Learning Russ Salakhutdinov Department of Computer Science! Department of Statistical Sciences! rsalakhu@cs.toronto.edu! h0p://www.cs.utoronto.ca/~rsalakhu/ Lecture 7 Approximate
More informationMeasurement error as missing data: the case of epidemiologic assays. Roderick J. Little
Measurement error as missing data: the case of epidemiologic assays Roderick J. Little Outline Discuss two related calibration topics where classical methods are deficient (A) Limit of quantification methods
More informationEstimating the marginal likelihood with Integrated nested Laplace approximation (INLA)
Estimating the marginal likelihood with Integrated nested Laplace approximation (INLA) arxiv:1611.01450v1 [stat.co] 4 Nov 2016 Aliaksandr Hubin Department of Mathematics, University of Oslo and Geir Storvik
More informationStable Limit Laws for Marginal Probabilities from MCMC Streams: Acceleration of Convergence
Stable Limit Laws for Marginal Probabilities from MCMC Streams: Acceleration of Convergence Robert L. Wolpert Institute of Statistics and Decision Sciences Duke University, Durham NC 778-5 - Revised April,
More informationStatistical Methods for Handling Incomplete Data Chapter 2: Likelihood-based approach
Statistical Methods for Handling Incomplete Data Chapter 2: Likelihood-based approach Jae-Kwang Kim Department of Statistics, Iowa State University Outline 1 Introduction 2 Observed likelihood 3 Mean Score
More informationLecture Notes based on Koop (2003) Bayesian Econometrics
Lecture Notes based on Koop (2003) Bayesian Econometrics A.Colin Cameron University of California - Davis November 15, 2005 1. CH.1: Introduction The concepts below are the essential concepts used throughout
More informationBayesian Statistical Methods. Jeff Gill. Department of Political Science, University of Florida
Bayesian Statistical Methods Jeff Gill Department of Political Science, University of Florida 234 Anderson Hall, PO Box 117325, Gainesville, FL 32611-7325 Voice: 352-392-0262x272, Fax: 352-392-8127, Email:
More informationBayesian Analysis of Latent Variable Models using Mplus
Bayesian Analysis of Latent Variable Models using Mplus Tihomir Asparouhov and Bengt Muthén Version 2 June 29, 2010 1 1 Introduction In this paper we describe some of the modeling possibilities that are
More informationσ(a) = a N (x; 0, 1 2 ) dx. σ(a) = Φ(a) =
Until now we have always worked with likelihoods and prior distributions that were conjugate to each other, allowing the computation of the posterior distribution to be done in closed form. Unfortunately,
More informationDefault Priors and Effcient Posterior Computation in Bayesian
Default Priors and Effcient Posterior Computation in Bayesian Factor Analysis January 16, 2010 Presented by Eric Wang, Duke University Background and Motivation A Brief Review of Parameter Expansion Literature
More information6. Fractional Imputation in Survey Sampling
6. Fractional Imputation in Survey Sampling 1 Introduction Consider a finite population of N units identified by a set of indices U = {1, 2,, N} with N known. Associated with each unit i in the population
More informationStat 542: Item Response Theory Modeling Using The Extended Rank Likelihood
Stat 542: Item Response Theory Modeling Using The Extended Rank Likelihood Jonathan Gruhl March 18, 2010 1 Introduction Researchers commonly apply item response theory (IRT) models to binary and ordinal
More informationIntroduction to Bayesian Methods
Introduction to Bayesian Methods Jessi Cisewski Department of Statistics Yale University Sagan Summer Workshop 2016 Our goal: introduction to Bayesian methods Likelihoods Priors: conjugate priors, non-informative
More informationSTAT Advanced Bayesian Inference
1 / 32 STAT 625 - Advanced Bayesian Inference Meng Li Department of Statistics Jan 23, 218 The Dirichlet distribution 2 / 32 θ Dirichlet(a 1,...,a k ) with density p(θ 1,θ 2,...,θ k ) = k j=1 Γ(a j) Γ(
More informationA Bayesian Nonparametric Approach to Monotone Missing Data in Longitudinal Studies with Informative Missingness
A Bayesian Nonparametric Approach to Monotone Missing Data in Longitudinal Studies with Informative Missingness A. Linero and M. Daniels UF, UT-Austin SRC 2014, Galveston, TX 1 Background 2 Working model
More informationHierarchical models. Dr. Jarad Niemi. August 31, Iowa State University. Jarad Niemi (Iowa State) Hierarchical models August 31, / 31
Hierarchical models Dr. Jarad Niemi Iowa State University August 31, 2017 Jarad Niemi (Iowa State) Hierarchical models August 31, 2017 1 / 31 Normal hierarchical model Let Y ig N(θ g, σ 2 ) for i = 1,...,
More informationMarkov chain Monte Carlo
1 / 26 Markov chain Monte Carlo Timothy Hanson 1 and Alejandro Jara 2 1 Division of Biostatistics, University of Minnesota, USA 2 Department of Statistics, Universidad de Concepción, Chile IAP-Workshop
More informationSpring 2006: Introduction to Markov Chain Monte Carlo (MCMC)
36-724 Spring 2006: Introduction to Marov Chain Monte Carlo (MCMC) Brian Juner February 16, 2006 Hierarchical Normal Model Direct Simulation An Alternative Approach: MCMC Complete Conditionals for Hierarchical
More informationDavid Giles Bayesian Econometrics
9. Model Selection - Theory David Giles Bayesian Econometrics One nice feature of the Bayesian analysis is that we can apply it to drawing inferences about entire models, not just parameters. Can't do
More informationBayesian Inference and MCMC
Bayesian Inference and MCMC Aryan Arbabi Partly based on MCMC slides from CSC412 Fall 2018 1 / 18 Bayesian Inference - Motivation Consider we have a data set D = {x 1,..., x n }. E.g each x i can be the
More informationApproximate Bayesian Computation
Approximate Bayesian Computation Michael Gutmann https://sites.google.com/site/michaelgutmann University of Helsinki and Aalto University 1st December 2015 Content Two parts: 1. The basics of approximate
More informationPrinciples of Bayesian Inference
Principles of Bayesian Inference Sudipto Banerjee and Andrew O. Finley 2 Biostatistics, School of Public Health, University of Minnesota, Minneapolis, Minnesota, U.S.A. 2 Department of Forestry & Department
More informationLecture 13 Fundamentals of Bayesian Inference
Lecture 13 Fundamentals of Bayesian Inference Dennis Sun Stats 253 August 11, 2014 Outline of Lecture 1 Bayesian Models 2 Modeling Correlations Using Bayes 3 The Universal Algorithm 4 BUGS 5 Wrapping Up
More informationStat 451 Lecture Notes Markov Chain Monte Carlo. Ryan Martin UIC
Stat 451 Lecture Notes 07 12 Markov Chain Monte Carlo Ryan Martin UIC www.math.uic.edu/~rgmartin 1 Based on Chapters 8 9 in Givens & Hoeting, Chapters 25 27 in Lange 2 Updated: April 4, 2016 1 / 42 Outline
More informationPrinciples of Bayesian Inference
Principles of Bayesian Inference Sudipto Banerjee 1 and Andrew O. Finley 2 1 Biostatistics, School of Public Health, University of Minnesota, Minneapolis, Minnesota, U.S.A. 2 Department of Forestry & Department
More informationEmbedding Supernova Cosmology into a Bayesian Hierarchical Model
1 / 41 Embedding Supernova Cosmology into a Bayesian Hierarchical Model Xiyun Jiao Statistic Section Department of Mathematics Imperial College London Joint work with David van Dyk, Roberto Trotta & Hikmatali
More informationA note on multiple imputation for general purpose estimation
A note on multiple imputation for general purpose estimation Shu Yang Jae Kwang Kim SSC meeting June 16, 2015 Shu Yang, Jae Kwang Kim Multiple Imputation June 16, 2015 1 / 32 Introduction Basic Setup Assume
More informationContents. Part I: Fundamentals of Bayesian Inference 1
Contents Preface xiii Part I: Fundamentals of Bayesian Inference 1 1 Probability and inference 3 1.1 The three steps of Bayesian data analysis 3 1.2 General notation for statistical inference 4 1.3 Bayesian
More informationNested Sampling. Brendon J. Brewer. brewer/ Department of Statistics The University of Auckland
Department of Statistics The University of Auckland https://www.stat.auckland.ac.nz/ brewer/ is a Monte Carlo method (not necessarily MCMC) that was introduced by John Skilling in 2004. It is very popular
More informationBayesian Phylogenetics:
Bayesian Phylogenetics: an introduction Marc A. Suchard msuchard@ucla.edu UCLA Who is this man? How sure are you? The one true tree? Methods we ve learned so far try to find a single tree that best describes
More informationMCMC for big data. Geir Storvik. BigInsight lunch - May Geir Storvik MCMC for big data BigInsight lunch - May / 17
MCMC for big data Geir Storvik BigInsight lunch - May 2 2018 Geir Storvik MCMC for big data BigInsight lunch - May 2 2018 1 / 17 Outline Why ordinary MCMC is not scalable Different approaches for making
More informationEmpirical Bayes Unfolding of Elementary Particle Spectra at the Large Hadron Collider
Empirical Bayes Unfolding of Elementary Particle Spectra at the Large Hadron Collider Mikael Kuusela Institute of Mathematics, EPFL Statistics Seminar, University of Bristol June 13, 2014 Joint work with
More informationBAYESIAN MODEL CRITICISM
Monte via Chib s BAYESIAN MODEL CRITICM Hedibert Freitas Lopes The University of Chicago Booth School of Business 5807 South Woodlawn Avenue, Chicago, IL 60637 http://faculty.chicagobooth.edu/hedibert.lopes
More informationTime Delay Estimation: Microlensing
Time Delay Estimation: Microlensing Hyungsuk Tak Stat310 15 Sep 2015 Joint work with Kaisey Mandel, David A. van Dyk, Vinay L. Kashyap, Xiao-Li Meng, and Aneta Siemiginowska Introduction Image Credit:
More informationST 740: Markov Chain Monte Carlo
ST 740: Markov Chain Monte Carlo Alyson Wilson Department of Statistics North Carolina State University October 14, 2012 A. Wilson (NCSU Stsatistics) MCMC October 14, 2012 1 / 20 Convergence Diagnostics:
More informationBayesian model selection for computer model validation via mixture model estimation
Bayesian model selection for computer model validation via mixture model estimation Kaniav Kamary ATER, CNAM Joint work with É. Parent, P. Barbillon, M. Keller and N. Bousquet Outline Computer model validation
More informationHastings-within-Gibbs Algorithm: Introduction and Application on Hierarchical Model
UNIVERSITY OF TEXAS AT SAN ANTONIO Hastings-within-Gibbs Algorithm: Introduction and Application on Hierarchical Model Liang Jing April 2010 1 1 ABSTRACT In this paper, common MCMC algorithms are introduced
More informationVariational Scoring of Graphical Model Structures
Variational Scoring of Graphical Model Structures Matthew J. Beal Work with Zoubin Ghahramani & Carl Rasmussen, Toronto. 15th September 2003 Overview Bayesian model selection Approximations using Variational
More informationNew Results of Fully Bayesian
UCI April 3, 2012 Calibration Samples Principle Component Analysis Model Building Three source parameter sampling schemes Simulation Quasar data sets Speed Up Pragmatic Bayesian Method Frequency Analysis
More informationFall 2017 STAT 532 Homework Peter Hoff. 1. Let P be a probability measure on a collection of sets A.
1. Let P be a probability measure on a collection of sets A. (a) For each n N, let H n be a set in A such that H n H n+1. Show that P (H n ) monotonically converges to P ( k=1 H k) as n. (b) For each n
More informationUsing Bayes Factors for Model Selection in High-Energy Astrophysics
Using Bayes Factors for Model Selection in High-Energy Astrophysics Shandong Zhao Department of Statistic, UCI April, 203 Model Comparison in Astrophysics Nested models (line detection in spectral analysis):
More informationParametric fractional imputation for missing data analysis
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 Biometrika (????),??,?, pp. 1 15 C???? Biometrika Trust Printed in
More informationStats 579 Intermediate Bayesian Modeling. Assignment # 2 Solutions
Stats 579 Intermediate Bayesian Modeling Assignment # 2 Solutions 1. Let w Gy) with y a vector having density fy θ) and G having a differentiable inverse function. Find the density of w in general and
More informationMarkov Chain Monte Carlo Methods
Markov Chain Monte Carlo Methods John Geweke University of Iowa, USA 2005 Institute on Computational Economics University of Chicago - Argonne National Laboaratories July 22, 2005 The problem p (θ, ω I)
More informationDS-GA 1003: Machine Learning and Computational Statistics Homework 7: Bayesian Modeling
DS-GA 1003: Machine Learning and Computational Statistics Homework 7: Bayesian Modeling Due: Tuesday, May 10, 2016, at 6pm (Submit via NYU Classes) Instructions: Your answers to the questions below, including
More informationBayesian Regression Linear and Logistic Regression
When we want more than point estimates Bayesian Regression Linear and Logistic Regression Nicole Beckage Ordinary Least Squares Regression and Lasso Regression return only point estimates But what if we
More informationModel Selection for Galaxy Shapes
Model Selection for Galaxy Shapes Chunzhe Zhang Advisor: Thomas C.M. Lee Collaborators: Vinay Kashyap, Andreas Zezas May 24, 2016 Chunzhe Zhang (UCDavis) Model Selection May 24, 2016 1 / 20 Overview 1
More informationMarkov Chain Monte Carlo
Department of Statistics The University of Auckland https://www.stat.auckland.ac.nz/~brewer/ Emphasis I will try to emphasise the underlying ideas of the methods. I will not be teaching specific software
More informationFoundations of Statistical Inference
Foundations of Statistical Inference Julien Berestycki Department of Statistics University of Oxford MT 2016 Julien Berestycki (University of Oxford) SB2a MT 2016 1 / 32 Lecture 14 : Variational Bayes
More informationAnalysis of Middle Censored Data with Exponential Lifetime Distributions
Analysis of Middle Censored Data with Exponential Lifetime Distributions Srikanth K. Iyer S. Rao Jammalamadaka Debasis Kundu Abstract Recently Jammalamadaka and Mangalam (2003) introduced a general censoring
More informationBayesian Networks in Educational Assessment
Bayesian Networks in Educational Assessment Estimating Parameters with MCMC Bayesian Inference: Expanding Our Context Roy Levy Arizona State University Roy.Levy@asu.edu 2017 Roy Levy MCMC 1 MCMC 2 Posterior
More informationShu Yang and Jae Kwang Kim. Harvard University and Iowa State University
Statistica Sinica 27 (2017), 000-000 doi:https://doi.org/10.5705/ss.202016.0155 DISCUSSION: DISSECTING MULTIPLE IMPUTATION FROM A MULTI-PHASE INFERENCE PERSPECTIVE: WHAT HAPPENS WHEN GOD S, IMPUTER S AND
More informationBayes Model Selection with Path Sampling: Factor Models
with Path Sampling: Factor Models Ritabrata Dutta and Jayanta K Ghosh Purdue University 07/02/11 Factor Models in Applications Factor Models in Applications Factor Models Factor Models and Factor analysis
More informationBayesian Linear Regression
Bayesian Linear Regression Sudipto Banerjee 1 Biostatistics, School of Public Health, University of Minnesota, Minneapolis, Minnesota, U.S.A. September 15, 2010 1 Linear regression models: a Bayesian perspective
More informationSTA 216, GLM, Lecture 16. October 29, 2007
STA 216, GLM, Lecture 16 October 29, 2007 Efficient Posterior Computation in Factor Models Underlying Normal Models Generalized Latent Trait Models Formulation Genetic Epidemiology Illustration Structural
More informationA Bayesian multi-dimensional couple-based latent risk model for infertility
A Bayesian multi-dimensional couple-based latent risk model for infertility Zhen Chen, Ph.D. Eunice Kennedy Shriver National Institute of Child Health and Human Development National Institutes of Health
More informationUsing Model Selection and Prior Specification to Improve Regime-switching Asset Simulations
Using Model Selection and Prior Specification to Improve Regime-switching Asset Simulations Brian M. Hartman, PhD ASA Assistant Professor of Actuarial Science University of Connecticut BYU Statistics Department
More informationFully Bayesian Analysis of Low-Count Astronomical Images
Fully Bayesian Analysis of Low-Count Astronomical Images 1 Alanna Connors 2 1 Department of Statistics University of California, Irvine 2 Eurika Scientific Thanks to James Chiang, Adam Roy, and The California
More informationMarkov Chain Monte Carlo (MCMC)
Markov Chain Monte Carlo (MCMC Dependent Sampling Suppose we wish to sample from a density π, and we can evaluate π as a function but have no means to directly generate a sample. Rejection sampling can
More informationHierarchical Bayesian Modeling
Hierarchical Bayesian Modeling Making scientific inferences about a population based on many individuals Angie Wolfgang NSF Postdoctoral Fellow, Penn State Astronomical Populations Once we discover an
More informationBayesian inference J. Daunizeau
Bayesian inference J. Daunizeau Brain and Spine Institute, Paris, France Wellcome Trust Centre for Neuroimaging, London, UK Overview of the talk 1 Probabilistic modelling and representation of uncertainty
More informationFirst Year Examination Department of Statistics, University of Florida
First Year Examination Department of Statistics, University of Florida August 19, 010, 8:00 am - 1:00 noon Instructions: 1. You have four hours to answer questions in this examination.. You must show your
More informationLecture : 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 informationEmbedding Astronomical Computer Models into Complex Statistical Models. David A. van Dyk
Embedding Astronomical Computer Models into Comple Statistical Models Statistics Section, Imperial College London UCLA, February 2012 Comple Analyses with Computer Models in Astronomy Outline Computer
More information