Michael Dewar

Point process modelling of the Afghan War Diary

Modern conflicts are characterized by an ever increasing use of information and sensing technology, resulting in vast amounts of high resolution data. Modelling and prediction of conflict, however, remain challenging tasks due to the heterogeneous and dynamic nature of the data typically available. Here we propose the use of dynamic spatiotemporal modelling tools for the identification of complex underlying processes in conflict, such as diffusion, relocation, heterogeneous escalation, and volatility. Using ideas from statistics, signal processing, and ecology, we provide a predictive framework able to assimilate data and give confidence estimates on the predictions. We demonstrate our methods on the WikiLeaks Afghan War Diary. Our results show that the approach allows deeper insights into conflict dynamics and allows a strikingly statistically accurate forward prediction of armed opposition group activity in 2010, based solely on data from previous years.

Tracing the Trajectory of Skill Learning With a Very Large Sample of Online Game Players

In the present study, we analyzed data from a very large sample (N = 854,064) of players of an online game involving rapid perception, decision making, and motor responding. Use of game data allowed us to connect, for the first time, rich details of training history with measures of performance from participants engaged for a sustained amount of time in effortful practice. We showed that lawful relations exist between practice amount and subsequent performance, and between practice spacing and subsequent performance. Our methodology allowed an in situ confirmation of results long established in the experimental literature on skill acquisition. Additionally, we showed that greater initial variation in performance is linked to higher subsequent performance, a result we link to the exploration/exploitation trade-off from the computational framework of reinforcement learning. We discuss the benefits and opportunities of behavioral data sets with very large sample sizes and suggest that this approach could be particularly fecund for studies of skill acquisition.

A data-driven framework for neural field modeling.

This paper presents a framework for creating neural field models from electrophysiological data. The Wilson and Cowan or Amari style neural field equations are used to form a parametric model, where the parameters are estimated from data. To illustrate the estimation framework, data is generated using the neural field equations incorporating modeled sensors enabling a comparison between the estimated and true parameters. To facilitate state and parameter estimation, we introduce a method to reduce the continuum neural field model using a basis function decomposition to form a finite-dimensional state-space model. Spatial frequency analysis methods are introduced that systematically specify the basis function configuration required to capture the dominant characteristics of the neural field. The estimation procedure consists of a two-stage iterative algorithm incorporating the unscented Rauch-Tung-Striebel smoother for state estimation and a least squares algorithm for parameter estimation. The results show that it is theoretically possible to reconstruct the neural field and estimate intracortical connectivity structure and synaptic dynamics with the proposed framework.

Inference in Hidden Markov Models with Explicit State Duration Distributions

In this letter, we borrow from the inference techniques developed for unbounded state-cardinality (nonparametric) variants of the HMM and use them to develop a tuning-parameter free, black-box inference procedure for explicit-state-duration hidden Markov models (EDHMM). EDHMMs are HMMs that have latent states consisting of both discrete state-indicator and discrete state-duration random variables. In contrast to the implicit geometric state duration distribution possessed by the standard HMM, EDHMMs allow the direct parameterization and estimation of per-state duration distributions. As most duration distributions are defined over the positive integers, truncation or other approximations are usually required to perform EDHMM inference.

Parameter estimation and inference for stochastic reaction-diffusion systems: application to morphogenesis in D. melanogaster

BackgroundReaction-diffusion systems are frequently used in systems biology to model developmental and signalling processes. In many applications, count numbers of the diffusing molecular species are very low, leading to the need to explicitly model the inherent variability using stochastic methods. Despite their importance and frequent use, parameter estimation for both deterministic and stochastic reaction-diffusion systems is still a challenging problem.ResultsWe present a Bayesian inference approach to solve both the parameter and state estimation problem for stochastic reaction-diffusion systems. This allows a determination of the full posterior distribution of the parameters (expected values and uncertainty). We benchmark the method by illustrating it on a simple synthetic experiment. We then test the method on real data about the diffusion of the morphogen Bicoid in Drosophila melanogaster. The results show how the precision with which parameters can be inferred varies dramatically, indicating that the ability to infer full posterior distributions on the parameters can have important experimental design consequences.ConclusionsThe results obtained demonstrate the feasibility and potential advantages of applying a Bayesian approach to parameter estimation in stochastic reaction-diffusion systems. In particular, the ability to estimate credibility intervals associated with parameter estimates can be precious for experimental design. Further work, however, will be needed to ensure the method can scale up to larger problems.

Data-Driven Spatio-Temporal Modeling Using the Integro-Difference Equation

A continuous-in-space, discrete-in-time dynamic spatio-temporal model known as the integro-difference equation (IDE) model is presented in the context of data-driven modeling. A novel decomposition of the IDE is derived, leading to state-space representation that does not couple the number of states with the number of observation locations or the number of parameters. Based on this state-space model, an expectation-maximization (EM) algorithm is developed in order to jointly estimate the IDE models spatial field and spatial mixing kernel. The resulting modeling framework is demonstrated on a set of examples.

Estimation and Model Selection for an IDE-Based Spatio-Temporal Model

A state space model of the stochastic spatio-temporal integro-difference equation (IDE) is derived. Based on multidimensional sampling theory, the dimensions of the state space and parameter space of the model are identified from the spatial bandwidth of the system and the support of the redistribution kernel of the IDE. When both the bandwidth and the kernel support are unknown, a method to propose a number of state space and parameter space dimensions is presented. These chosen dimensions result in a number of candidate model structures. Bayesian model selection, making use of Bayes factor, the data augmentation algorithm and importance sampling, is then used to identify the model best suited to represent the data in a maximum a posteriori sense.

A Canonical Space-Time State Space Model: State and Parameter Estimation

The maximum likelihood estimation of a dynamic spatiotemporal model is introduced, centred around the inclusion of a prior arbitrary spatiotemporal neighborhood description. The neighborhood description defines a specific parameterization of the state transition matrix, chosen on the basis of prior knowledge about the system. The model used is inspired by the spatiotemporal ARMA (STARMA) model, but the representation used is based on the standard state-space model. The inclusion of the neighborhood into an expectation-maximization based joint state and parameter estimation algorithm allows for accurate characterization of the spatiotemporal model. The process of including the neighborhood, and the effect it has on the maximum likelihood parameter estimate is described and demonstrated in this paper.

Spatiotemporal multi-resolution approximation of the Amari type neural field model

Neural fields are spatially continuous state variables described by integro-differential equations, which are well suited to describe the spatiotemporal evolution of cortical activations on multiple scales. Here we develop a multi-resolution approximation (MRA) framework for the integro-difference equation (IDE) neural field model based on semi-orthogonal cardinal B-spline wavelets. In this way, a flexible framework is created, whereby both macroscopic and microscopic behavior of the system can be represented simultaneously. State and parameter estimation is performed using the expectation maximization (EM) algorithm. A synthetic example is provided to demonstrate the framework.

Identifying Hosts of Families of Viruses: A Machine Learning Approach

Identifying emerging viral pathogens and characterizing their transmission is essential to developing effective public health measures in response to an epidemic. Phylogenetics, though currently the most popular tool used to characterize the likely host of a virus, can be ambiguous when studying species very distant to known species and when there is very little reliable sequence information available in the early stages of the outbreak of disease. Motivated by an existing framework for representing biological sequence information, we learn sparse, tree-structured models, built from decision rules based on subsequences, to predict viral hosts from protein sequence data using popular discriminative machine learning tools. Furthermore, the predictive motifs robustly selected by the learning algorithm are found to show strong host-specificity and occur in highly conserved regions of the viral proteome.