On the Quotient Function Employed in the Blind Source Separation Problem

MSC 2010: 42C40, 94A12

The association between glycated haemoglobin levels and P100 visual evoked potentials in diabetes mellitus

Diabetes mellitus (DM) is a metabolic disorder which is characterised by hyperglycaemia resulting from defects in insulin secretion, insulin action or both. The long-term specific effects of DM include the development of retinopathy, nephropathy and neuropathy. Cardiac disease, peripheral arterial and cerebrovascular disease are also known to be linked with DM. Type 1 diabetes mellitus (T1DM) accounts for approximately 10% of all individuals with DM, and insulin therapy is the only available treatment. Type 2 diabetes mellitus (T2DM) accounts for 90% of all individuals with DM. Diet, exercise, oral hypoglycaemic agents and occasionally exogenous insulin are used to manage T2DM. The diagnosis of DM is made where the glycated haemoglobin (HbA1c) percentage is greater than 6.5%. Pattern-reversal visual evoked potential (PVEP) testing is an objective means of evaluating impulse conduction along the central nervous pathways. Increased peak time of the visual P100 waveform is an expression of structural damage at the level of myelinated optic nerve fibres. This was an observational cross sectional study. The participants were grouped into two phases. Phase 1, the control group, consisted of 30 healthy non-diabetic participants. Phase 2 comprised of 104 diabetic participants of whom 52 had an HbA1c greater than 10% (poorly controlled DM) and 52 whose HbA1c was 10% and less (moderately controlled DM). The aim of this study was to firstly observe the possible association between glycated haemoglobin levels and P100 peak time of pattern-reversal visual evoked potentials (PVEPs) in DM. Secondly, to assess whether the central nervous system (CNS) and in particular visual function is affected by type and/or duration of DM. The cut-off values to define P100 peak time delay was calculated as the mean P100 peak time plus 2.5 X standard deviations as measured for the non-diabetic control group, and were 110.64 ms for the right eye. The proportion of delayed P100 peak time amounted to 38.5% for both diabetic groups, thus the poorly controlled group (HbA1c > 10%) did not pose an increased risk for delayed P100 peak time, relative to the moderately controlled group (HbA1c ≤ 10%). The P100 PVEP results for this study, do however, reflect significant delay (p < 0.001) of the DM group as compared to the non-diabetic group; thus, subclincal neuropathy of the CNS occurs in 38.5% of cases. The duration of DM and type of DM had no influence on the P100 peak time measurements.

Motion correction algorithm of lung tumors for respiratory gated PET images

Respiratory gating in lung PET imaging to compensate for respiratory motion artifacts is a current research issue with broad potential impact on quantitation, diagnosis and clinical management of lung tumors. However, PET images collected at discrete bins can be significantly affected by noise as there are lower activity counts in each gated bin unless the total PET acquisition time is prolonged, so that gating methods should be combined with imaging-based motion correction and registration methods. The aim of this study was to develop and validate a fast and practical solution to the problem of respiratory motion for the detection and accurate quantitation of lung tumors in PET images. This included: (1) developing a computer-assisted algorithm for PET/CT images that automatically segments lung regions in CT images, identifies and localizes lung tumors of PET images; (2) developing and comparing different registration algorithms which processes all the information within the entire respiratory cycle and integrate all the tumor in different gated bins into a single reference bin. Four registration/integration algorithms: Centroid Based, Intensity Based, Rigid Body and Optical Flow registration were compared as well as two registration schemes: Direct Scheme and Successive Scheme. Validation was demonstrated by conducting experiments with the computerized 4D NCAT phantom and with a dynamic lung-chest phantom imaged using a GE PET/CT System. Iterations were conducted on different size simulated tumors and different noise levels. Static tumors without respiratory motion were used as gold standard; quantitative results were compared with respect to tumor activity concentration, cross-correlation coefficient, relative noise level and computation time. Comparing the results of the tumors before and after correction, the tumor activity values and tumor volumes were closer to the static tumors (gold standard). Higher correlation values and lower noise were also achieved after applying the correction algorithms. With this method the compromise between short PET scan time and reduced image noise can be achieved, while quantification and clinical analysis become fast and precise.

An improved framework for dynamic origin-destination (O-D) matrix estimation

This dissertation aims to improve the performance of existing assignment-based dynamic origin-destination (O-D) matrix estimation models to successfully apply Intelligent Transportation Systems (ITS) strategies for the purposes of traffic congestion relief and dynamic traffic assignment (DTA) in transportation network modeling. The methodology framework has two advantages over the existing assignment-based dynamic O-D matrix estimation models. First, it combines an initial O-D estimation model into the estimation process to provide a high confidence level of initial input for the dynamic O-D estimation model, which has the potential to improve the final estimation results and reduce the associated computation time. Second, the proposed methodology framework can automatically convert traffic volume deviation to traffic density deviation in the objective function under congested traffic conditions. Traffic density is a better indicator for traffic demand than traffic volume under congested traffic condition, thus the conversion can contribute to improving the estimation performance. The proposed method indicates a better performance than a typical assignment-based estimation model (Zhou et al., 2003) in several case studies. In the case study for I-95 in Miami-Dade County, Florida, the proposed method produces a good result in seven iterations, with a root mean square percentage error (RMSPE) of 0.010 for traffic volume and a RMSPE of 0.283 for speed. In contrast, Zhou's model requires 50 iterations to obtain a RMSPE of 0.023 for volume and a RMSPE of 0.285 for speed. In the case study for Jacksonville, Florida, the proposed method reaches a convergent solution in 16 iterations with a RMSPE of 0.045 for volume and a RMSPE of 0.110 for speed, while Zhou's model needs 10 iterations to obtain the best solution, with a RMSPE of 0.168 for volume and a RMSPE of 0.179 for speed. The successful application of the proposed methodology framework to real road networks demonstrates its ability to provide results both with satisfactory accuracy and within a reasonable time, thus establishing its potential usefulness to support dynamic traffic assignment modeling, ITS systems, and other strategies.

Trajectory privacy preservation in mobile Wireless Sensor Networks

In recent years, there has been an enormous growth of location-aware devices, such as GPS embedded cell phones, mobile sensors and radio-frequency identification tags. The age of combining sensing, processing and communication in one device, gives rise to a vast number of applications leading to endless possibilities and a realization of mobile Wireless Sensor Network (mWSN) applications. As computing, sensing and communication become more ubiquitous, trajectory privacy becomes a critical piece of information and an important factor for commercial success. While on the move, sensor nodes continuously transmit data streams of sensed values and spatiotemporal information, known as ``trajectory information". If adversaries can intercept this information, they can monitor the trajectory path and capture the location of the source node. ^ This research stems from the recognition that the wide applicability of mWSNs will remain elusive unless a trajectory privacy preservation mechanism is developed. The outcome seeks to lay a firm foundation in the field of trajectory privacy preservation in mWSNs against external and internal trajectory privacy attacks. First, to prevent external attacks, we particularly investigated a context-based trajectory privacy-aware routing protocol to prevent the eavesdropping attack. Traditional shortest-path oriented routing algorithms give adversaries the possibility to locate the target node in a certain area. We designed the novel privacy-aware routing phase and utilized the trajectory dissimilarity between mobile nodes to mislead adversaries about the location where the message started its journey. Second, to detect internal attacks, we developed a software-based attestation solution to detect compromised nodes. We created the dynamic attestation node chain among neighboring nodes to examine the memory checksum of suspicious nodes. The computation time for memory traversal had been improved compared to the previous work. Finally, we revisited the trust issue in trajectory privacy preservation mechanism designs. We used Bayesian game theory to model and analyze cooperative, selfish and malicious nodes' behaviors in trajectory privacy preservation activities.^

Design and development of a one-degree-of-freedom force-reflecting manual controller prototype for teleoperation

The present research is carried out from the viewpoint of primarily space applications where human lives may be in danger if they are to work under these conditions. This work proposes to develop a one-degree-of-freedom (1-DOF) force-reflecting manual controller (FRMC) prototype for teleoperation, and address the effects of time delays commonly found in space applications where the control is accomplished via the earth-based control stations. To test the FRMC, a mobile robot (PPRK) and a slider-bar were developed and integrated to the 1-DOF FRMC. The software developed in Visual Basic is able to telecontrol any platform that uses an SV203 controller through the internet and it allows the remote system to send feedback information which may be in the form of visual or force signals. Time delay experiments were conducted on the platform and the effects of time delay on the FRMC system operation have been studied and delineated.

Motion Correction Algorithm of Lung Tumors for Respiratory Gated PET Images

Respiratory gating in lung PET imaging to compensate for respiratory motion artifacts is a current research issue with broad potential impact on quantitation, diagnosis and clinical management of lung tumors. However, PET images collected at discrete bins can be significantly affected by noise as there are lower activity counts in each gated bin unless the total PET acquisition time is prolonged, so that gating methods should be combined with imaging-based motion correction and registration methods. The aim of this study was to develop and validate a fast and practical solution to the problem of respiratory motion for the detection and accurate quantitation of lung tumors in PET images. This included: (1) developing a computer-assisted algorithm for PET/CT images that automatically segments lung regions in CT images, identifies and localizes lung tumors of PET images; (2) developing and comparing different registration algorithms which processes all the information within the entire respiratory cycle and integrate all the tumor in different gated bins into a single reference bin. Four registration/integration algorithms: Centroid Based, Intensity Based, Rigid Body and Optical Flow registration were compared as well as two registration schemes: Direct Scheme and Successive Scheme. Validation was demonstrated by conducting experiments with the computerized 4D NCAT phantom and with a dynamic lung-chest phantom imaged using a GE PET/CT System. Iterations were conducted on different size simulated tumors and different noise levels. Static tumors without respiratory motion were used as gold standard; quantitative results were compared with respect to tumor activity concentration, cross-correlation coefficient, relative noise level and computation time. Comparing the results of the tumors before and after correction, the tumor activity values and tumor volumes were closer to the static tumors (gold standard). Higher correlation values and lower noise were also achieved after applying the correction algorithms. With this method the compromise between short PET scan time and reduced image noise can be achieved, while quantification and clinical analysis become fast and precise.

Characterization of dynamic speckle sequences using principal component analysis and image descriptors

Speckle is being used as a characterization tool for the analysis of the dynamic of slow varying phenomena occurring in biological and industrial samples. The retrieved data takes the form of a sequence of speckle images. The analysis of these images should reveal the inner dynamic of the biological or physical process taking place in the sample. Very recently, it has been shown that principal component analysis is able to split the original data set in a collection of classes. These classes can be related with the dynamic of the observed phenomena. At the same time, statistical descriptors of biospeckle images have been used to retrieve information on the characteristics of the sample. These statistical descriptors can be calculated in almost real time and provide a fast monitoring of the sample. On the other hand, principal component analysis requires longer computation time but the results contain more information related with spatial-temporal pattern that can be identified with physical process. This contribution merges both descriptions and uses principal component analysis as a pre-processing tool to obtain a collection of filtered images where a simpler statistical descriptor can be calculated. The method has been applied to slow-varying biological and industrial processes

Bayesian Inference in Large-scale Problems

Many modern applications fall into the category of "large-scale" statistical problems, in which both the number of observations n and the number of features or parameters p may be large. Many existing methods focus on point estimation, despite the continued relevance of uncertainty quantification in the sciences, where the number of parameters to estimate often exceeds the sample size, despite huge increases in the value of n typically seen in many fields. Thus, the tendency in some areas of industry to dispense with traditional statistical analysis on the basis that "n=all" is of little relevance outside of certain narrow applications. The main result of the Big Data revolution in most fields has instead been to make computation much harder without reducing the importance of uncertainty quantification. Bayesian methods excel at uncertainty quantification, but often scale poorly relative to alternatives. This conflict between the statistical advantages of Bayesian procedures and their substantial computational disadvantages is perhaps the greatest challenge facing modern Bayesian statistics, and is the primary motivation for the work presented here.

Two general strategies for scaling Bayesian inference are considered. The first is the development of methods that lend themselves to faster computation, and the second is design and characterization of computational algorithms that scale better in n or p. In the first instance, the focus is on joint inference outside of the standard problem of multivariate continuous data that has been a major focus of previous theoretical work in this area. In the second area, we pursue strategies for improving the speed of Markov chain Monte Carlo algorithms, and characterizing their performance in large-scale settings. Throughout, the focus is on rigorous theoretical evaluation combined with empirical demonstrations of performance and concordance with the theory.

One topic we consider is modeling the joint distribution of multivariate categorical data, often summarized in a contingency table. Contingency table analysis routinely relies on log-linear models, with latent structure analysis providing a common alternative. Latent structure models lead to a reduced rank tensor factorization of the probability mass function for multivariate categorical data, while log-linear models achieve dimensionality reduction through sparsity. Little is known about the relationship between these notions of dimensionality reduction in the two paradigms. In Chapter 2, we derive several results relating the support of a log-linear model to nonnegative ranks of the associated probability tensor. Motivated by these findings, we propose a new collapsed Tucker class of tensor decompositions, which bridge existing PARAFAC and Tucker decompositions, providing a more flexible framework for parsimoniously characterizing multivariate categorical data. Taking a Bayesian approach to inference, we illustrate empirical advantages of the new decompositions.

Latent class models for the joint distribution of multivariate categorical, such as the PARAFAC decomposition, data play an important role in the analysis of population structure. In this context, the number of latent classes is interpreted as the number of genetically distinct subpopulations of an organism, an important factor in the analysis of evolutionary processes and conservation status. Existing methods focus on point estimates of the number of subpopulations, and lack robust uncertainty quantification. Moreover, whether the number of latent classes in these models is even an identified parameter is an open question. In Chapter 3, we show that when the model is properly specified, the correct number of subpopulations can be recovered almost surely. We then propose an alternative method for estimating the number of latent subpopulations that provides good quantification of uncertainty, and provide a simple procedure for verifying that the proposed method is consistent for the number of subpopulations. The performance of the model in estimating the number of subpopulations and other common population structure inference problems is assessed in simulations and a real data application.

In contingency table analysis, sparse data is frequently encountered for even modest numbers of variables, resulting in non-existence of maximum likelihood estimates. A common solution is to obtain regularized estimates of the parameters of a log-linear model. Bayesian methods provide a coherent approach to regularization, but are often computationally intensive. Conjugate priors ease computational demands, but the conjugate Diaconis--Ylvisaker priors for the parameters of log-linear models do not give rise to closed form credible regions, complicating posterior inference. In Chapter 4 we derive the optimal Gaussian approximation to the posterior for log-linear models with Diaconis--Ylvisaker priors, and provide convergence rate and finite-sample bounds for the Kullback-Leibler divergence between the exact posterior and the optimal Gaussian approximation. We demonstrate empirically in simulations and a real data application that the approximation is highly accurate, even in relatively small samples. The proposed approximation provides a computationally scalable and principled approach to regularized estimation and approximate Bayesian inference for log-linear models.

Another challenging and somewhat non-standard joint modeling problem is inference on tail dependence in stochastic processes. In applications where extreme dependence is of interest, data are almost always time-indexed. Existing methods for inference and modeling in this setting often cluster extreme events or choose window sizes with the goal of preserving temporal information. In Chapter 5, we propose an alternative paradigm for inference on tail dependence in stochastic processes with arbitrary temporal dependence structure in the extremes, based on the idea that the information on strength of tail dependence and the temporal structure in this dependence are both encoded in waiting times between exceedances of high thresholds. We construct a class of time-indexed stochastic processes with tail dependence obtained by endowing the support points in de Haan's spectral representation of max-stable processes with velocities and lifetimes. We extend Smith's model to these max-stable velocity processes and obtain the distribution of waiting times between extreme events at multiple locations. Motivated by this result, a new definition of tail dependence is proposed that is a function of the distribution of waiting times between threshold exceedances, and an inferential framework is constructed for estimating the strength of extremal dependence and quantifying uncertainty in this paradigm. The method is applied to climatological, financial, and electrophysiology data.

The remainder of this thesis focuses on posterior computation by Markov chain Monte Carlo. The Markov Chain Monte Carlo method is the dominant paradigm for posterior computation in Bayesian analysis. It has long been common to control computation time by making approximations to the Markov transition kernel. Comparatively little attention has been paid to convergence and estimation error in these approximating Markov Chains. In Chapter 6, we propose a framework for assessing when to use approximations in MCMC algorithms, and how much error in the transition kernel should be tolerated to obtain optimal estimation performance with respect to a specified loss function and computational budget. The results require only ergodicity of the exact kernel and control of the kernel approximation accuracy. The theoretical framework is applied to approximations based on random subsets of data, low-rank approximations of Gaussian processes, and a novel approximating Markov chain for discrete mixture models.

Data augmentation Gibbs samplers are arguably the most popular class of algorithm for approximately sampling from the posterior distribution for the parameters of generalized linear models. The truncated Normal and Polya-Gamma data augmentation samplers are standard examples for probit and logit links, respectively. Motivated by an important problem in quantitative advertising, in Chapter 7 we consider the application of these algorithms to modeling rare events. We show that when the sample size is large but the observed number of successes is small, these data augmentation samplers mix very slowly, with a spectral gap that converges to zero at a rate at least proportional to the reciprocal of the square root of the sample size up to a log factor. In simulation studies, moderate sample sizes result in high autocorrelations and small effective sample sizes. Similar empirical results are observed for related data augmentation samplers for multinomial logit and probit models. When applied to a real quantitative advertising dataset, the data augmentation samplers mix very poorly. Conversely, Hamiltonian Monte Carlo and a type of independence chain Metropolis algorithm show good mixing on the same dataset.

Distributed Feature Selection in Large n and Large p Regression Problems

Fitting statistical models is computationally challenging when the sample size or the dimension of the dataset is huge. An attractive approach for down-scaling the problem size is to first partition the dataset into subsets and then fit using distributed algorithms. The dataset can be partitioned either horizontally (in the sample space) or vertically (in the feature space), and the challenge arise in defining an algorithm with low communication, theoretical guarantees and excellent practical performance in general settings. For sample space partitioning, I propose a MEdian Selection Subset AGgregation Estimator ({\em message}) algorithm for solving these issues. The algorithm applies feature selection in parallel for each subset using regularized regression or Bayesian variable selection method, calculates the `median' feature inclusion index, estimates coefficients for the selected features in parallel for each subset, and then averages these estimates. The algorithm is simple, involves very minimal communication, scales efficiently in sample size, and has theoretical guarantees. I provide extensive experiments to show excellent performance in feature selection, estimation, prediction, and computation time relative to usual competitors.

While sample space partitioning is useful in handling datasets with large sample size, feature space partitioning is more effective when the data dimension is high. Existing methods for partitioning features, however, are either vulnerable to high correlations or inefficient in reducing the model dimension. In the thesis, I propose a new embarrassingly parallel framework named {\em DECO} for distributed variable selection and parameter estimation. In {\em DECO}, variables are first partitioned and allocated to m distributed workers. The decorrelated subset data within each worker are then fitted via any algorithm designed for high-dimensional problems. We show that by incorporating the decorrelation step, DECO can achieve consistent variable selection and parameter estimation on each subset with (almost) no assumptions. In addition, the convergence rate is nearly minimax optimal for both sparse and weakly sparse models and does NOT depend on the partition number m. Extensive numerical experiments are provided to illustrate the performance of the new framework.

For datasets with both large sample sizes and high dimensionality, I propose a new "divided-and-conquer" framework {\em DEME} (DECO-message) by leveraging both the {\em DECO} and the {\em message} algorithm. The new framework first partitions the dataset in the sample space into row cubes using {\em message} and then partition the feature space of the cubes using {\em DECO}. This procedure is equivalent to partitioning the original data matrix into multiple small blocks, each with a feasible size that can be stored and fitted in a computer in parallel. The results are then synthezied via the {\em DECO} and {\em message} algorithm in a reverse order to produce the final output. The whole framework is extremely scalable.

Results of wave modelling for Moreton Bay, Southeast Queensland, and a reef lagoon off Lizard Island, Great Barrier Reef, Australia

The distribution, abundance, behaviour, and morphology of marine species is affected by spatial variability in the wave environment. Maps of wave metrics (e.g. significant wave height Hs, peak energy wave period Tp, and benthic wave orbital velocity URMS) are therefore useful for predictive ecological models of marine species and ecosystems. A number of techniques are available to generate maps of wave metrics, with varying levels of complexity in terms of input data requirements, operator knowledge, and computation time. Relatively simple "fetch-based" models are generated using geographic information system (GIS) layers of bathymetry and dominant wind speed and direction. More complex, but computationally expensive, "process-based" models are generated using numerical models such as the Simulating Waves Nearshore (SWAN) model. We generated maps of wave metrics based on both fetch-based and process-based models and asked whether predictive performance in models of benthic marine habitats differed. Predictive models of seagrass distribution for Moreton Bay, Southeast Queensland, and Lizard Island, Great Barrier Reef, Australia, were generated using maps based on each type of wave model. For Lizard Island, performance of the process-based wave maps was significantly better for describing the presence of seagrass, based on Hs, Tp, and URMS. Conversely, for the predictive model of seagrass in Moreton Bay, based on benthic light availability and Hs, there was no difference in performance using the maps of the different wave metrics. For predictive models where wave metrics are the dominant factor determining ecological processes it is recommended that process-based models be used. Our results suggest that for models where wave metrics provide secondarily useful information, either fetch- or process-based models may be equally useful.

A cloud-based system for measuring radiation treatment plan similarity

PURPOSE: Radiation therapy is used to treat cancer using carefully designed plans that maximize the radiation dose delivered to the target and minimize damage to healthy tissue, with the dose administered over multiple occasions. Creating treatment plans is a laborious process and presents an obstacle to more frequent replanning, which remains an unsolved problem. However, in between new plans being created, the patient's anatomy can change due to multiple factors including reduction in tumor size and loss of weight, which results in poorer patient outcomes. Cloud computing is a newer technology that is slowly being used for medical applications with promising results. The objective of this work was to design and build a system that could analyze a database of previously created treatment plans, which are stored with their associated anatomical information in studies, to find the one with the most similar anatomy to a new patient. The analyses would be performed in parallel on the cloud to decrease the computation time of finding this plan. METHODS: The system used SlicerRT, a radiation therapy toolkit for the open-source platform 3D Slicer, for its tools to perform the similarity analysis algorithm. Amazon Web Services was used for the cloud instances on which the analyses were performed, as well as for storage of the radiation therapy studies and messaging between the instances and a master local computer. A module was built in SlicerRT to provide the user with an interface to direct the system on the cloud, as well as to perform other related tasks. RESULTS: The cloud-based system out-performed previous methods of conducting the similarity analyses in terms of time, as it analyzed 100 studies in approximately 13 minutes, and produced the same similarity values as those methods. It also scaled up to larger numbers of studies to analyze in the database with a small increase in computation time of just over 2 minutes. CONCLUSION: This system successfully analyzes a large database of radiation therapy studies and finds the one that is most similar to a new patient, which represents a potential step forward in achieving feasible adaptive radiation therapy replanning.

An efficient parallel MoM to analyze microstrip structures

[EN]The increasing use of microstrip technology require more accurate analysis methods like full wave method of moments. However, this involves a great computational effort. To reduce the computation time, an alternative parallel method to analyze irregular microstrip structures is presented in this paper. This method calculates the unknown surface current on the planar structure trough a irregular rectangular division using basis and weighted functions. The parallel algorithm performs the calculus of a [Z] matrix and then solves the system using current densities as the unknowns. This parallel program was implemented in the IBM-SP2 using MPI library.

Experimental study of forces and energies during shearing of steel sheet with angled tools

Shearing is a fast and inexpensive method to cut sheet metal that has been used since the beginning of the industrialism. Consequently, published experimental studies of shearing can be found from over a century back in time. Recent studies, however, are due to the availability of low cost digital computation power, mostly based on finite element simulations that guarantees quick results. Still, for validation of models and simulations, accurate experimental data is a requisite. When applicable, 2D models are in general desirable over 3D models because of advantages like low computation time and easy model formulation. Shearing of sheet metal with parallel tools is successfully modelled in 2D with a plane strain approximation, but with angled tools the approximation is less obvious. Therefore, plane strain approximations for shearing with angled tools were evaluated by shear experiments of high accuracy. Tool angle, tool clearance, and clamping of the sheet were varied in the experiments. The results showed that the measured forces in shearing with angled tools can be approximately calculated using force measurements from shearing with parallel tools. Shearing energy was introduced as a quantifiable measure of suitable tool clearance range. The effects of the shearing parameters on forces were in agreement with previous studies. Based on the agreement between calculations and experiments, analysis based on a plane strain assumption is considered applicable for angled tools with a small (up to 2 degrees) rake angle.

Dynamic load balancing of distributed memory parallel computational mechanics using unstructured meshes for multi-physical modelling

As the complexity of parallel applications increase, the performance limitations resulting from computational load imbalance become dominant. Mapping the problem space to the processors in a parallel machine in a manner that balances the workload of each processors will typically reduce the run-time. In many cases the computation time required for a given calculation cannot be predetermined even at run-time and so static partition of the problem returns poor performance. For problems in which the computational load across the discretisation is dynamic and inhomogeneous, for example multi-physics problems involving fluid and solid mechanics with phase changes, the workload for a static subdomain will change over the course of a computation and cannot be estimated beforehand. For such applications the mapping of loads to process is required to change dynamically, at run-time in order to maintain reasonable efficiency. The issue of dynamic load balancing are examined in the context of PHYSICA, a three dimensional unstructured mesh multi-physics continuum mechanics computational modelling code.