A comparison of variational and Markov chain Monte Carlo methods for inference in partially observed stochastic dynamic systems

In recent work we have developed a novel variational inference method for partially observed systems governed by stochastic differential equations. In this paper we provide a comparison of the Variational Gaussian Process Smoother with an exact solution computed using a Hybrid Monte Carlo approach to path sampling, applied to a stochastic double well potential model. It is demonstrated that the variational smoother provides us a very accurate estimate of mean path while conditional variance is slightly underestimated. We conclude with some remarks as to the advantages and disadvantages of the variational smoother. © 2008 Springer Science + Business Media LLC.

Variational Markov Chain Monte Carlo for Bayesian smoothing of non-linear niffusions

In this paper we develop set of novel Markov chain Monte Carlo algorithms for Bayesian smoothing of partially observed non-linear diffusion processes. The sampling algorithms developed herein use a deterministic approximation to the posterior distribution over paths as the proposal distribution for a mixture of an independence and a random walk sampler. The approximating distribution is sampled by simulating an optimized time-dependent linear diffusion process derived from the recently developed variational Gaussian process approximation method. Flexible blocking strategies are introduced to further improve mixing, and thus the efficiency, of the sampling algorithms. The algorithms are tested on two diffusion processes: one with double-well potential drift and another with SINE drift. The new algorithm's accuracy and efficiency is compared with state-of-the-art hybrid Monte Carlo based path sampling. It is shown that in practical, finite sample, applications the algorithm is accurate except in the presence of large observation errors and low observation densities, which lead to a multi-modal structure in the posterior distribution over paths. More importantly, the variational approximation assisted sampling algorithm outperforms hybrid Monte Carlo in terms of computational efficiency, except when the diffusion process is densely observed with small errors in which case both algorithms are equally efficient.

Variational Markov chain Monte Carlo for Bayesian smoothing of non-linear diffusions

In this paper we develop set of novel Markov Chain Monte Carlo algorithms for Bayesian smoothing of partially observed non-linear diffusion processes. The sampling algorithms developed herein use a deterministic approximation to the posterior distribution over paths as the proposal distribution for a mixture of an independence and a random walk sampler. The approximating distribution is sampled by simulating an optimized time-dependent linear diffusion process derived from the recently developed variational Gaussian process approximation method. The novel diffusion bridge proposal derived from the variational approximation allows the use of a flexible blocking strategy that further improves mixing, and thus the efficiency, of the sampling algorithms. The algorithms are tested on two diffusion processes: one with double-well potential drift and another with SINE drift. The new algorithm's accuracy and efficiency is compared with state-of-the-art hybrid Monte Carlo based path sampling. It is shown that in practical, finite sample applications the algorithm is accurate except in the presence of large observation errors and low to a multi-modal structure in the posterior distribution over paths. More importantly, the variational approximation assisted sampling algorithm outperforms hybrid Monte Carlo in terms of computational efficiency, except when the diffusion process is densely observed with small errors in which case both algorithms are equally efficient. © 2011 Springer-Verlag.

Resonance Energy Transfer-Based Molecular Switch Designed Using a Systematic Design Process Based on Monte Carlo Methods and Markov Chains

A RET network consists of a network of photo-active molecules called chromophores that can participate in inter-molecular energy transfer called resonance energy transfer (RET). RET networks are used in a variety of applications including cryptographic devices, storage systems, light harvesting complexes, biological sensors, and molecular rulers. In this dissertation, we focus on creating a RET device called closed-diffusive exciton valve (C-DEV) in which the input to output transfer function is controlled by an external energy source, similar to a semiconductor transistor like the MOSFET. Due to their biocompatibility, molecular devices like the C-DEVs can be used to introduce computing power in biological, organic, and aqueous environments such as living cells. Furthermore, the underlying physics in RET devices are stochastic in nature, making them suitable for stochastic computing in which true random distribution generation is critical.

In order to determine a valid configuration of chromophores for the C-DEV, we developed a systematic process based on user-guided design space pruning techniques and built-in simulation tools. We show that our C-DEV is 15x better than C-DEVs designed using ad hoc methods that rely on limited data from prior experiments. We also show ways in which the C-DEV can be improved further and how different varieties of C-DEVs can be combined to form more complex logic circuits. Moreover, the systematic design process can be used to search for valid chromophore network configurations for a variety of RET applications.

We also describe a feasibility study for a technique used to control the orientation of chromophores attached to DNA. Being able to control the orientation can expand the design space for RET networks because it provides another parameter to tune their collective behavior. While results showed limited control over orientation, the analysis required the development of a mathematical model that can be used to determine the distribution of dipoles in a given sample of chromophore constructs. The model can be used to evaluate the feasibility of other potential orientation control techniques.

Avaliação das exposições dos envolvidos em procedimentos intervencionistas usando método Monte Carlo

Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)

Population Monte Carlo Algorithm in High Dimensions

The population Monte Carlo algorithm is an iterative importance sampling scheme for solving static problems. We examine the population Monte Carlo algorithm in a simplified setting, a single step of the general algorithm, and study a fundamental problem that occurs in applying importance sampling to high-dimensional problem. The precision of the computed estimate from the simplified setting is measured by the asymptotic variance of estimate under conditions on the importance function. We demonstrate the exponential growth of the asymptotic variance with the dimension and show that the optimal covariance matrix for the importance function can be estimated in special cases.

A sequential Monte Carlo approach to design for population pharmacokinetics studies

Here we present a sequential Monte Carlo approach that can be used to find optimal designs. Our focus is on the design of phase III clinical trials where the derivation of sampling windows is required, along with the optimal sampling schedule. The search is conducted via a particle filter which traverses a sequence of target distributions artificially constructed via an annealed utility. The algorithm derives a catalogue of highly efficient designs which, not only contain the optimal, but can also be used to derive sampling windows. We demonstrate our approach by designing a hypothetical phase III clinical trial.

Experimental evaluation of MCDTK, the Monte Carlo DICOM Tool-Kit

The Monte Carlo DICOM Tool-Kit (MCDTK) is a software suite designed for treatment plan dose verification, using the BEAMnrc and DOSXYZnrc Monte Carlo codes. MCDTK converts DICOM-format treatment plan information into Monte Carlo input files and compares the results of Monte Carlo treatment simulations with conventional treatment planning dose calculations. In this study, a treatment is planned using a commercial treatment planning system, delivered to a pelvis phantom containing ten thermoluminescent dosimeters and simulated using BEAMnrc and DOSXYZnrc using inputs derived from MCDTK. The dosimetric accuracy of the Monte Carlo data is then evaluated via comparisons with the dose distribution obtained from the treatment planning system as well as the in-phantom point dose measurements. The simulated beam arrangement produced by MCDTK is found to be in geometric agreement with the planned treatment. An isodose display generated from the Monte Carlo data by MCDTK shows general agreement with the isodose display obtained from the treatment planning system, except for small regions around density heterogeneities in the phantom, where the pencil-beam dose calculation performed by the treatment planning systemis likely to be less accurate. All point dose measurements agree with the Monte Carlo data obtained using MCDTK, within confidence limits, and all except one of these point dose measurements show closer agreement with theMonte Carlo data than with the doses calculated by the treatment planning system. This study provides a simple demonstration of the geometric and dosimetric accuracy ofMonte Carlo simulations based on information from MCDTK.

A sequential Monte Carlo approach to the sequential design for discriminating between rival continuous data models

Here we present a sequential Monte Carlo approach to Bayesian sequential design for the incorporation of model uncertainty. The methodology is demonstrated through the development and implementation of two model discrimination utilities; mutual information and total separation, but it can also be applied more generally if one has different experimental aims. A sequential Monte Carlo algorithm is run for each rival model (in parallel), and provides a convenient estimate of the marginal likelihood (of each model) given the data, which can be used for model comparison and in the evaluation of utility functions. A major benefit of this approach is that it requires very little problem specific tuning and is also computationally efficient when compared to full Markov chain Monte Carlo approaches. This research is motivated by applications in drug development and chemical engineering.

Calculation of dose kernels for radiotherapy applications using the GEANT 4 Monte Carlo toolkit

Dose kernels may be used to calculate dose distributions in radiotherapy (as described by Ahnesjo et al., 1999). Their calculation requires use of Monte Carlo methods, usually by forcing interactions to occur at a point. The Geant4 Monte Carlo toolkit provides a capability to force interactions to occur in a particular volume. We have modified this capability and created a Geant4 application to calculate dose kernels in cartesian, cylindrical, and spherical scoring systems. The simulation considers monoenergetic photons incident at the origin of a 3 m x 3 x 9 3 m water volume. Photons interact via compton, photo-electric, pair production, and rayleigh scattering. By default, Geant4 models photon interactions by sampling a physical interaction length (PIL) for each process. The process returning the smallest PIL is then considered to occur. In order to force the interaction to occur within a given length, L_FIL, we scale each PIL according to the formula: PIL_forced = L_FIL 9 (1 - exp(-PIL/PILo)) where PILo is a constant. This ensures that the process occurs within L_FIL, whilst correctly modelling the relative probability of each process. Dose kernels were produced for an incident photon energy of 0.1, 1.0, and 10.0 MeV. In order to benchmark the code, dose kernels were also calculated using the EGSnrc Edknrc user code. Identical scoring systems were used; namely, the collapsed cone approach of the Edknrc code. Relative dose difference images were then produced. Preliminary results demonstrate the ability of the Geant4 application to reproduce the shape of the dose kernels; median relative dose differences of 12.6, 5.75, and 12.6 % were found for an incident photon energy of 0.1, 1.0, and 10.0 MeV respectively.

Monte Carlo-based diode design for correction-less small field dosimetry

Due to their small collecting volume diodes are commonly used in small field dosimetry. However the relative sensitivity of a diode increases with decreasing small field size. Conversely, small air gaps have been shown to cause a significant decrease in the sensitivity of a detector as the field size is decreased. Therefore this study uses Monte Carlo simulations to look at introducing air upstream to diodes such that they measure with a constant sensitivity across all field sizes in small field dosimetry. Varying thicknesses of air were introduced onto the upstream end of two commercial diodes (PTW 60016 photon diode and PTW 60017 electron diode), as well as a theoretical unenclosed silicon chip using field sizes as small as 5 mm × 5 mm . The metric D_(w,Q)/D_(Det,Q) used in this study represents the ratio of the dose to a point of water to the dose to the diode active volume, for a particular field size and location. The optimal thickness of air required to provide a constant sensitivity across all small field sizes was found by plotting D_(w,Q)/D_(Det,Q) as a function of introduced air gap size for various field sizes, and finding the intersection point of these plots. That is, the point at which D_(w,Q)/D_(Det,Q) was constant for all field sizes was found. The optimal thickness of air was calculated to be 3.3 mm, 1.15 mm and 0.10 mm for the photon diode, electron diode and unenclosed silicon chip respectively. The variation in these results was due to the different design of each detector. When calculated with the new diode design incorporating the upstream air gap, k_(Q_clin 〖,Q〗_msr)^(f_clin 〖,f〗_msr ) was equal to unity to within statistical uncertainty (0.5 %) for all three diodes. Cross-axis profile measurements were also improved with the new detector design. The upstream air gap could be implanted on the commercial diodes via a cap consisting of the air cavity surrounded by water equivalent material. The results for the unclosed silicon chip show that an ideal small field dosimetry diode could be created by using a silicon chip with a small amount of air above it.

Effect of CT electron density identification on Monte Carlo simulations of radiotherapeutic treatments

Introduction: The accurate identification of tissue electron densities is of great importance for Monte Carlo (MC) dose calculations. When converting patient CT data into a voxelised format suitable for MC simulations, however, it is common to simplify the assignment of electron densities so that the complex tissues existing in the human body are categorized into a few basic types. This study examines the effects that the assignment of tissue types and the calculation of densities can have on the results of MC simulations, for the particular case of a Siemen’s Sensation 4 CT scanner located in a radiotherapy centre where QA measurements are routinely made using 11 tissue types (plus air). Methods: DOSXYZnrc phantoms are generated from CT data, using the CTCREATE user code, with the relationship between Hounsfield units (HU) and density determined via linear interpolation between a series of specified points on the ‘CT-density ramp’ (see Figure 1(a)). Tissue types are assigned according to HU ranges. Each voxel in the DOSXYZnrc phantom therefore has an electron density (electrons/cm3) defined by the product of the mass density (from the HU conversion) and the intrinsic electron density (electrons /gram) (from the material assignment), in that voxel. In this study, we consider the problems of density conversion and material identification separately: the CT-density ramp is simplified by decreasing the number of points which define it from 12 down to 8, 3 and 2; and the material-type-assignment is varied by defining the materials which comprise our test phantom (a Supertech head) as two tissues and bone, two plastics and bone, water only and (as an extreme case) lead only. The effect of these parameters on radiological thickness maps derived from simulated portal images is investigated. Results & Discussion: Increasing the degree of simplification of the CT-density ramp results in an increasing effect on the resulting radiological thickness calculated for the Supertech head phantom. For instance, defining the CT-density ramp using 8 points, instead of 12, results in a maximum radiological thickness change of 0.2 cm, whereas defining the CT-density ramp using only 2 points results in a maximum radiological thickness change of 11.2 cm. Changing the definition of the materials comprising the phantom between water and plastic and tissue results in millimetre-scale changes to the resulting radiological thickness. When the entire phantom is defined as lead, this alteration changes the calculated radiological thickness by a maximum of 9.7 cm. Evidently, the simplification of the CT-density ramp has a greater effect on the resulting radiological thickness map than does the alteration of the assignment of tissue types. Conclusions: It is possible to alter the definitions of the tissue types comprising the phantom (or patient) without substantially altering the results of simulated portal images. However, these images are very sensitive to the accurate identification of the HU-density relationship. When converting data from a patient’s CT into a MC simulation phantom, therefore, all possible care should be taken to accurately reproduce the conversion between HU and mass density, for the specific CT scanner used. Acknowledgements: This work is funded by the NHMRC, through a project grant, and supported by the Queensland University of Technology (QUT) and the Royal Brisbane and Women's Hospital (RBWH), Brisbane, Australia. The authors are grateful to the staff of the RBWH, especially Darren Cassidy, for assistance in obtaining the phantom CT data used in this study. The authors also wish to thank Cathy Hargrave, of QUT, for assistance in formatting the CT data, using the Pinnacle TPS. Computational resources and services used in this work were provided by the HPC and Research Support Group, QUT, Brisbane, Australia.

Controlled Monte Carlo data generation for statistical damage identification employing Mahalanobis squared distance

The use of Mahalanobis squared distance–based novelty detection in statistical damage identification has become increasingly popular in recent years. The merit of the Mahalanobis squared distance–based method is that it is simple and requires low computational effort to enable the use of a higher dimensional damage-sensitive feature, which is generally more sensitive to structural changes. Mahalanobis squared distance–based damage identification is also believed to be one of the most suitable methods for modern sensing systems such as wireless sensors. Although possessing such advantages, this method is rather strict with the input requirement as it assumes the training data to be multivariate normal, which is not always available particularly at an early monitoring stage. As a consequence, it may result in an ill-conditioned training model with erroneous novelty detection and damage identification outcomes. To date, there appears to be no study on how to systematically cope with such practical issues especially in the context of a statistical damage identification problem. To address this need, this article proposes a controlled data generation scheme, which is based upon the Monte Carlo simulation methodology with the addition of several controlling and evaluation tools to assess the condition of output data. By evaluating the convergence of the data condition indices, the proposed scheme is able to determine the optimal setups for the data generation process and subsequently avoid unnecessarily excessive data. The efficacy of this scheme is demonstrated via applications to a benchmark structure data in the field.

Transdimensional sequential Monte Carlo using variational Bayes — SMCVB

A new transdimensional Sequential Monte Carlo (SMC) algorithm called SM- CVB is proposed. In an SMC approach, a weighted sample of particles is generated from a sequence of probability distributions which ‘converge’ to the target distribution of interest, in this case a Bayesian posterior distri- bution. The approach is based on the use of variational Bayes to propose new particles at each iteration of the SMCVB algorithm in order to target the posterior more efficiently. The variational-Bayes-generated proposals are not limited to a fixed dimension. This means that the weighted particle sets that arise can have varying dimensions thereby allowing us the option to also estimate an appropriate dimension for the model. This novel algorithm is outlined within the context of finite mixture model estimation. This pro- vides a less computationally demanding alternative to using reversible jump Markov chain Monte Carlo kernels within an SMC approach. We illustrate these ideas in a simulated data analysis and in applications.

Monte-Carlo estimation of time-dependent statistical characteristics of random dynamical systems

The problem of estimating the time-dependent statistical characteristics of a random dynamical system is studied under two different settings. In the first, the system dynamics is governed by a differential equation parameterized by a random parameter, while in the second, this is governed by a differential equation with an underlying parameter sequence characterized by a continuous time Markov chain. We propose, for the first time in the literature, stochastic approximation algorithms for estimating various time-dependent process characteristics of the system. In particular, we provide efficient estimators for quantities such as the mean, variance and distribution of the process at any given time as well as the joint distribution and the autocorrelation coefficient at different times. A novel aspect of our approach is that we assume that information on the parameter model (i.e., its distribution in the first case and transition probabilities of the Markov chain in the second) is not available in either case. This is unlike most other work in the literature that assumes availability of such information. Also, most of the prior work in the literature is geared towards analyzing the steady-state system behavior of the random dynamical system while our focus is on analyzing the time-dependent statistical characteristics which are in general difficult to obtain. We prove the almost sure convergence of our stochastic approximation scheme in each case to the true value of the quantity being estimated. We provide a general class of strongly consistent estimators for the aforementioned statistical quantities with regular sample average estimators being a specific instance of these. We also present an application of the proposed scheme on a widely used model in population biology. Numerical experiments in this framework show that the time-dependent process characteristics as obtained using our algorithm in each case exhibit excellent agreement with exact results. (C) 2010 Elsevier Inc. All rights reserved.