arXiv subjects

Hans C. Eggers

Publications and source records attributed to Hans C. Eggers.

10 recordsLinked to original sources

arXiv2019

Stochastic Gradient Annealed Importance Sampling for Efficient Online Marginal Likelihood Estimation

We consider estimating the marginal likelihood in settings with independent and identically distributed (i.i.d.) data. We propose estimating the predictive distributions in a sequential factorization of the marginal likelihood in such settings by using stochastic gradient Markov Chain Monte Carlo techniques. This approach is far more efficient than traditional marginal likelihood estimation techniques such as nested sampling and annealed importance sampling due to its use of mini-batches to approximate the likelihood. Stability of the estimates is provided by an adaptive annealing schedule. The resulting stochastic gradient annealed importance sampling (SGAIS) technique, which is the key contribution of our paper, enables us to estimate the marginal likelihood of a number of models considerably faster than traditional approaches, with no noticeable loss of accuracy. An important benefit of our approach is that the marginal likelihood is calculated in an online fashion as data becomes available, allowing the estimates to be used for applications such as online weighted model combination.

stat.ML↗

arXiv2018

Bayesian modelling of uncertainties of Monte Carlo radiative-transfer simulations

One of the big challenges in astrophysics is the comparison of complex simulations to observations. As many codes do not directly generate observables (e.g. hydrodynamic simulations), the last step in the modelling process is often a radiative-transfer treatment. For this step, the community relies increasingly on Monte Carlo radiative transfer due to the ease of implementation and scalability with computing power. We consider simulations in which the number of photon packets is Poisson distributed, while the weight assigned to a single photon packet follows any distribution of choice. We show how to estimate the statistical uncertainty of the sum of weights in each bin from the output of a single radiative-transfer simulation. Our Bayesian approach produces a posterior distribution that is valid for any number of packets in a bin, even zero packets, and is easy to implement in practice. Our analytic results for large number of packets show that we generalise existing methods that are valid only in limiting cases. The statistical problem considered here appears in identical form in a wide range of Monte Carlo simulations including particle physics and importance sampling. It is particularly powerful in extracting information when the available data are sparse or quantities are small.

astro-ph.IM↗

arXiv2015

Optimal modeling of 1D azimuth correlations in the context of Bayesian inference

Analysis and interpretation of spectrum and correlation data from high-energy nuclear collisions is currently controversial because two opposing physics narratives derive contradictory implications from the same data---one narrative claiming collision dynamics is dominated by dijet production and projectile-nucleon fragmentation, the other claiming collision dynamics is dominated by a dense, flowing QCD medium. Opposing interpretations seem to be supported by alternative data models, and current model-comparison schemes are unable to distinguish between them. There is clearly need for a convincing new methodology to break the deadlock. In this study we introduce Bayesian Inference (BI) methods applied to angular correlation data as a basis to evaluate competing data models. For simplicity the data considered are projections of 2D angular correlations onto 1D azimuth from three centrality classes of 200 GeV AuAu collisions. We consider several data models typical of current model choices, including Fourier series (FS) and a Gaussian plus various combinations of individual cosine components. We evaluate model performance with BI methods and with power-spectrum (PS) analysis. We find that the FS-only model is rejected in all cases by Bayesian analysis which always prefers a Gaussian. A cylindrical quadrupole cos(2ϕ) is required in some cases but rejected for 0-5%-central AuAu collisions. Given a Gaussian centered at the azimuth origin "higher harmonics" cos(mϕ) for m > 2 are rejected. A model consisting of Gaussian + dipole cos(ϕ) + quadrupole cos(2ϕ) provides good 1D data descriptions in all cases.

hep-ph↗

arXiv2015

Bayesian model comparison for one-dimensional azimuthal correlations in 200GeV AuAu collisions

In the context of data modeling and comparisons between different fit models, Bayesian analysis calls that model best which has the largest evidence, the prior-weighted integral over model parameters of the likelihood function. Evidence calculations automatically take into account both the usual chi-squared measure and an Occam factor which quantifies the price for adding extra parameters. Applying Bayesian analysis to projections onto azimuth of 2D angular correlations from 200 GeV AuAu collisions, we consider typical model choices including Fourier series and a Gaussian plus combinations of individual cosine components. We find that models including a Gaussian component are consistently preferred over pure Fourier-series parametrizations, sometimes strongly so. For 0-5% central collisions the Gaussian-plus-dipole model performs better than Fourier Series models or any other combination of Gaussian-plus-multipoles.

hep-ph↗

arXiv2012

From chi-squared to Bayesian model comparison and Levy expansions of Bose-Einstein correlations in e+e- reactions

The usual chi-squared method of fit quality assessment is a special case of the more general method of Bayesian model comparison which involves integrals of the likelihood and prior over all possible values of all parameters. We introduce new parametrisations based on systematic expansions around the stretched exponential or Fourier-transformed Lévy source distribution, and utilise the increased discriminating power of the Bayesian approach to evaluate the relative probability of these models to be true representations of a recently measured Bose-Einstein correlation data in e+e- annihilations at LEP.

nucl-th↗

arXiv2003

Stochastic energy-cascade model for 1+1 dimensional fully developed turbulence

Geometrical random multiplicative cascade processes are often used to model positive-valued multifractal fields such as the energy dissipation in fully developed turbulence. We propose a dynamical generalization describing the energy dissipation in terms of a continuous and homogeneous stochastic field in one space and one time dimension. In the model, correlations originate in the overlap of the respective spacetime histories of field amplitudes. The theoretical two- and three-point correlation functions are found to be in good agreement with their equal-time counterparts extracted from wind tunnel turbulent shear flow data.

cond-mat↗

arXiv2001

Dynamical random multiplicative cascade model in 1+1 dimensions

Geometrical random multiplicative cascade processes are often used to model positive-valued multifractal fields such as for example the energy dissipation field of fully developed turbulence. A dynamical generalisation of these models is proposed, which describes the continuous and homogeneous stochastic evolution of the field in one space and one time dimension. Two-point correlation functions are calculated.

cond-mat.stat-mech↗

arXiv2001

Translationally invariant cumulants in energy cascade models of turbulence

In the context of random multiplicative energy cascade processes, we derive analytical expressions for translationally invariant one- and two-point cumulants in logarithmic field amplitudes. Such cumulants make it possible to distinguish between hitherto equally successful cascade generator models and hence supplement lowest-order multifractal scaling exponents and multiplier distributions.

nlin.CD↗

arXiv1998

Spatial correlations of singularity strengths in multifractal branching processes

The n-point statistics of singularity strength variables for multiplicative branching processes is calculated from an analytic expression of the corresponding multivariate generating function. The key ingredient is a branching generating function which can be understood as a natural generalisation of the multifractal mass exponents. Various random multiplicative cascade processes pertaining to fully developed turbulence are discussed.

chao-dyn↗

arXiv1998

Analytic multivariate generating function for random multiplicative cascade processes

We have found an analytic expression for the multivariate generating function governing all n-point statistics of random multiplicative cascade processes. The variable appropriate for this generating function is the logarithm of the energy density, ln epsilon, rather than epsilon itself. All cumulant statistics become sums over derivatives of ``branching generating functions'' which are Laplace transforms of the splitting functions and completely determine the cascade process. We show that the branching generating function is a generalization of the multifractal mass exponents. Two simple models from fully developed turbulence illustrate the new formalism.

chao-dyn↗