Past Seminars

Page 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Structured population epidemic models
Frank G Ball
University of Nottingham
Start time: 3:15 pm
Date: Tuesday 18 July 2006
Location: Russell Love Theatre, Richard Berry Building, The University of Melbourne

Standard deterministic models of epidemics implicitly assume that the population among which the disease is spreading is locally as well as globally large, in the sense that if the population is partitioned into groups, e.g. by age, sex and/or geographical location, then each of these groups, and not just the total population, is large. The same assumption is made when analysing the large-population behaviour of many stochastic epidemic models. However, this assumption is unrealistic for many human epidemics, since such populations contain small social groups, such as households, school classes and workplaces, in which transmission is likely to be enhanced. Thus, there has been a growing interest in models for epidemics among populations whose structure remains locally finite as the population becomes large. This talk is concerned with a general class of structured-population epidemic models, in which individuals mix at two levels: global and local. After some introductory comments, a stochastic model for SIR (susceptible - infected - removed) epidemics among a closed finite population is described, in which during its infectious period a typical infective makes both local and global contacts. Each local contact of a given infective is with an individual chosen independently according to a contact distribution centred on that infective and each global contact is with an individual chosen independently and uniformly from the whole population. The threshold behaviour of the model is determined, as is the asymptotic final outcome in the event of an epidemic taking off. The theory is specialised to (i) the households model, in which the population is partitioned into households and local contacts are chosen uniformly within an infective's household; (ii) the overlapping groups model, in which the population is partitioned in several ways, with local uniform mixing within the elements of the partitions; and (iii) the great circle model, in which individuals are equally spaced on a circle and local contacts are nearest-neighbour. A main use of epidemic models is to evaluate the efficacy of control measures, such as vaccination, and optimal vaccination policies for the households model are also considered.

---

Info-Gap Forecasting and the Advantage of Sub-Optimal Models
Professor Yakov Ben-Haim
Technion - Israel Institute of Technology
Start time: 3:15 pm
Date: Friday 14 July 2006
Location: Theatre 1, Old Geology Building, The University of Melbourne

We consider forecasting in linear discrete-time systems. Historical data indicate that the transition matrix is constant over time. However, the underlying laws governing the system are unknown and it is uncertain that the system properties will remain constant. An info-gap model is used to represent uncertainty in the future transition matrix. The forecaster desires the average forecast of a specific state variable to be within a specified interval around the correct value. Traditionally, forecasting is based on using a model which has optimal fidelity to historical data. Our first theorem asserts the existence, and indicates the construction, of models with sub-optimal fidelity to historical data which are more robust to model error than the optimal model. Our second theorem identifies conditions in which the probability of forecast success increases with increasing robustness to model error. Combining these results leads to a methodology for identifying reliable forecasting models for systems whose trajectories evolve in unknown ways over time.

---

Algorithmic differentiation for analysis of global change and the carbon cycle
Prof. Ian Enting
MASCOS, University of Melbourne
Start time: 3:15 pm
Date: Friday 23 June 2006
Location: AMSI Seminar Room, Ground Floor, 111 Barry Street, Carlton

Numerical models of natural systems are often expressed as sets of differential equations (DEs), generally first-order in time. Commonly 'the model' refers to a computer implementation that integrates these DEs. However many of the processes involved in analysing models involve differentiation: sensitivity analysis, calibration by minimising a cost-function based on data-mismatch, etc. For large numerical models, it is convenient to have tools that generate such derivatives directly from the 'model' source code. This talk describes the use of operator overloading in C++ and Fortran-90 to transform a model into a tangent linear model that integrates DEs corresponding to parameter sensitivities.

---

Penalised spline support vector classifiers
Matt Wand
The University of NSW
Start time: 3:15 pm
Date: Friday 16 June 2006
Location: Russell Love Theatre, Richard Berry Building, The University of Melbourne

Support vector machines are an elegant and often effective means of building a classifier. They are defined up to choice of a kernel function and some auxiliary parameters. However, for many popular kernels, support vector classifiers can suffer from poor scalability for large training sample sizes; lack of interpretability and the curse of dimensionality. This talk describes kernels based on the notion of penalised splines (from nonparametric regression) and explains how they can alleviate the aforementioned problems.

---

Cycles and patterns in permutations, lattice paths, and exclusion processes
Robert Parviainen
The University of Melbourne
Start time: 3:15 pm
Date: Friday 9 June 2006
Location: Theatre 1, Ground Floor, 111 Barry Street, Carlton

Recently a lattice path representation of the stationary distributions for some exclusion processes was derived.  This enables us to write the generating function for the stationary distributions as a continued fraction. These results in turn may be used to show deep connections between these exclusion processes and permutation statistics. More specifically, parameters in the stationary distribution generating function correspond to, for example,  a) the number of cycles, and b) the number of occurrences of a certain pattern in permutations. This bi-statistic of cycles and patterns in permutations appears, despite high interest in patterns in permutations, to be previously unstudied.

---

Page 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15