BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Eric Stone (ANU)
DTSTART:20220510T060000Z
DTEND:20220510T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/1
DESCRIPTION:Title: Stories of mathematics and computational science in genetic mapping
\nby Eric Stone (ANU) as part of ANU Mathematics and Computational Sci
ences Seminar\n\nLecture held in Room 1.33\, Hanna Neumann Building #145.\
n\nAbstract\nTraits of interest often vary within a population\, leading b
iologists to investigate the genetic basis of that observed variation. Thi
s can be done directly via an association study\, in which one of many met
hods is used to identify correlational patterns that link genetic variatio
n to trait variation. Alternatively\, in experimental systems\, individual
s can be selectively bred to create a “genetic mapping population” wit
h a more desirable signal-to-noise ratio. In this talk\, I will share my e
xperience creating mapping populations as a vehicle to introducing some of
the mathematical and computational challenges that have ensued. I will di
scuss combinational and probabilistic issues that arise in ideal populatio
ns\, contrasted by some algorithmic concerns that arise in natural populat
ions. My goal is to provide a sampling of accessible problems in mathemati
cs and computational science encountered in a practical biological context
.\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/1/
END:VEVENT
BEGIN:VEVENT
SUMMARY:David Ham (Imperial College London)
DTSTART:20220524T060000Z
DTEND:20220524T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/2
DESCRIPTION:Title: Automating forward and inverse finite element simulation in Firedra
ke and Dolfin-adjoint\nby David Ham (Imperial College London) as part
of ANU Mathematics and Computational Sciences Seminar\n\nLecture held in R
oom 1.33\, Hanna Neumann Building #145.\n\nAbstract\nSimulating continuous
systems modelled by PDEs underpins much of computational science and engi
neering. Each simulation is a complex combination of PDEs\, parametrisatio
ns\, discretisations\, preconditioners and solvers. The precise combinatio
n that is optimal is different for each application and changes with the h
ardware\, or as further advances in numerical mathematics are made. Many (
possibly most) simulation challenges in science and engineering are actual
ly inverse problems in which parameters are sought\, sensitivities analyse
d and/or data assimilated.\n\nHere I will present Firedrake\, an automated
system for generating numerical solutions to PDEs from a high level mathe
matical specification. I will examine some of the capabilities of the syst
em before lifting the lid on the sequence of automated mathematical transf
ormations that make it possible. I will also cover the interaction with do
lfin-adjoint to produce gradients of solution functionals by solving the a
djoint PDE.\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/2/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Paul Malcolm (ANU DST)
DTSTART:20220607T060000Z
DTEND:20220607T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/3
DESCRIPTION:Title: New representations for a semi-Markov chain and related filters
\nby Paul Malcolm (ANU DST) as part of ANU Mathematics and Computational S
ciences Seminar\n\nLecture held in Room 1.33\, Hanna Neumann Building #145
.\n\nAbstract\nIt is now usual that the null-hypothesis for a finite-state
stochastic process is conveniently taken to be the standard Markov chain.
In the absence of any other system knowledge this is the model that is of
ten used. Some reasons for this are\; Markov chains are relatively simple\
, they have been well studied and much is known about these processes. Add
ed to this there are now decades of history applying the standard Hidden M
arkov Model (HMM) to: defence science\, gene sequencing\, health science\,
machine learning\, artificial intelligence and many other areas. In this
seminar we will briefly recall two common application domains of estimatio
n with latent Markov processes\, 1) parts-of-speech tagging (POS) in natur
al language processing and 2) tracking a maneuvering object with a Jump Ma
rkov System. Semi-Markov models relax an implicit feature of every state i
n a first-order time-homogeneous Markov chains\, that is\, the sojourn ran
dom variables of such states are geometrically distributed and are therefo
re\, (uniquely) memoryless random variables. In contrast\, semi-Markov cha
ins allow arbitrary sojourn models. Consequently\, a Hidden semi-Markov Mo
del (HsMM) offers a richer class of model\, but retains the classical HMM
as a special degenerate case.\n\nThe main task we address in this seminar
concerns model calibration\, or parameter estimation of a HsMM. We develop
an Expectation Maximization (EM) algorithm to compute the best fitting (i
n the Maximum Likelihood sense) HsMM for a given set of observation data.
There are several parts to this task\, the first is to derive a recursive
filter and smoother for a partially observed semi-Markov chain. The second
and more challenging part of the task is to derive filters and smoothers
for various processes derived from the latent semi-Markov chain\, for exam
ple\, a counting process that counts the number of transitions between two
distinct states labelled "i" and "j"\, up to an including time k. We will
see that estimators for such quantities are non-trivial\, largely because
of the sojourn dependence in transition probabilities.\n\nThe estimators
we present are all for partially observed joint events\, that is\, the sta
te of the semi-Markov chain at time "k" and the cumulative time it has rem
ained in this state. This means we are assured of exponential forgetting o
f initial conditions in our estimators. Separate estimators for individual
quantities such as the semi-Markov state alone are easily computed via ma
rginalization.\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/3/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quanling Deng (ANU)
DTSTART:20220802T060000Z
DTEND:20220802T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/4
DESCRIPTION:Title: Superparameterisation of Arctic sea ice floes\nby Quanling Deng
(ANU) as part of ANU Mathematics and Computational Sciences Seminar\n\nLe
cture held in Room 1.33\, Hanna Neumann Building #145.\n\nAbstract\nIn thi
s talk\, I will start with some quick facts about Arctic sea ice floes and
then give a quick review of the evolution of sea ice models. The first mo
dels are Eulerian continuum models that describe the sea ice floes as visc
ous-plastics (Hilber 1979). Lagrangian particle models have been developed
recently\, showing improved model performance\, especially in ice-margina
l zones where sea ice is fragmented. The most successful one is the discre
te element method (DEM). It characterises the physical quantities of each
sea ice floe along its trajectory under the Lagrangian coordinates. The ma
jor challenges are 1) model coupling in different frames of reference (Lag
rangian for sea ice while Eulerian for the ocean and atmosphere dynamics)\
; 2) the heavy computational cost when the number of the floes is large\;
and 3) inaccurate floe parameterisation when the floe distribution has mul
tiscale features. In this talk\, I will present a superfloe parameterisati
on to reduce the computational cost and a superparameterisation to capture
the multiscale features. The superfloe parameterisation algorithm generat
es a small number of superfloes that effectively approximate a considerabl
e number of the floes. The parameterisation scheme satisfies several impor
tant physics constraints that guarantee similar short-term dynamical behav
iour while maintaining long-range uncertainties\, especially the non-Gauss
ian statistical features\, of the full system. In addition\, the superfloe
parameterisation facilitates noise inflation in data assimilation that re
covers the unobserved ocean field underneath the sea ice. To capture the m
ultiscale features\, we follow the derivation of the Boltzmann equation fo
r particles and superparameterise the sea ice floes as continuity equation
s governing the statistical moments of mass density and linear and angular
velocities. This leads to a particle-continuum coupled model. The continu
um part captures the large scales and the particle part captures the small
scales. The particle model is localised and fully parallelised for comput
ation efficiency. I will present several numerical experiments to demonstr
ate the success of the proposed schemes. This is joint work with Nan Chen
(UW-Madison) and Sam Stechmann (UW-Madison).\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/4/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Minh Bui (ANU)
DTSTART:20220705T060000Z
DTEND:20220705T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/5
DESCRIPTION:Title: Phylogenetic inference in the genomic era\nby Minh Bui (ANU) as
part of ANU Mathematics and Computational Sciences Seminar\n\nLecture hel
d in Room 1.33\, Hanna Neumann Building #145.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/5/
END:VEVENT
BEGIN:VEVENT
SUMMARY:John Taylor (CSIRO)
DTSTART:20220712T060000Z
DTEND:20220712T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/6
DESCRIPTION:Title: TBC\nby John Taylor (CSIRO) as part of ANU Mathematics and Comp
utational Sciences Seminar\n\nLecture held in Room 1.33\, Hanna Neumann Bu
ilding #145.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/6/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Yu Lin (ANU)
DTSTART:20220719T060000Z
DTEND:20220719T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/7
DESCRIPTION:Title: Solving Genome Puzzles\nby Yu Lin (ANU) as part of ANU Mathemat
ics and Computational Sciences Seminar\n\nLecture held in Room 1.33\, Hann
a Neumann Building #145.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/7/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Giuseppe Barca (ANU)
DTSTART:20220816T060000Z
DTEND:20220816T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/8
DESCRIPTION:Title: Towards Exascale Computational Quantum Mechanics\nby Giuseppe B
arca (ANU) as part of ANU Mathematics and Computational Sciences Seminar\n
\nLecture held in Room 1.33\, Hanna Neumann Building #145.\nAbstract: TBA\
n
LOCATION:https://stable.researchseminars.org/talk/anumacs/8/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Linda Stals (ANU)
DTSTART:20220913T060000Z
DTEND:20220913T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/10
DESCRIPTION:Title: Fault Tolerant Iterative Solvers\nby Linda Stals (ANU) as part
of ANU Mathematics and Computational Sciences Seminar\n\nLecture held in
Room 1.33\, Hanna Neumann Building #145.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/10/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Vanessa Robins (ANU)
DTSTART:20220927T060000Z
DTEND:20220927T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/11
DESCRIPTION:Title: Topological Data Analysis\nby Vanessa Robins (ANU) as part of
ANU Mathematics and Computational Sciences Seminar\n\nLecture held in Room
1.33\, Hanna Neumann Building #145.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/11/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Charles O'Neill & Jack Miller (ANU)
DTSTART:20220823T060000Z
DTEND:20220823T070000Z
DTSTAMP:20260404T094833Z
UID:anumacs/12
DESCRIPTION:Title: Eigenvalue initialisation and regularisation for koopman autoencod
ers and beyond\nby Charles O'Neill & Jack Miller (ANU) as part of ANU
Mathematics and Computational Sciences Seminar\n\nLecture held in Room 1.3
3\, Hanna Neumann Building #145.\n\nAbstract\nRecent efforts have been mad
e to learn the Koopman operator with predictive autoencoders. However\, li
ttle attention has been payed to the initialisation of these networks. Not
ing the importance of eigenvalues to the action of a linear operator\, one
may ask whether it would be useful to employ them in the initialisation a
nd regularisation of these autoencoders? To answer this\, we devise a spec
tral eigenvalue initialisation and eigenvalue penalty scheme. Having done
so\, we discover that eigenvalues do in fact have great utility for this p
urpose. We demonstrate that in learning a Koopman operator for a damped dr
iven pendulum\, appropriate initialisation and regularisation can improve
initial performance by an order of magnitude. We also show with this syste
m that as the dissipative element of a dynamical system decreases\, the ut
ility of unit circle initialisation schemes increase and the utility of di
fferent regularisation schemes change. Additionally\, we show that the ben
efits of eigenvalue initialisation and regularisation generalise to real-w
orld cyclone wind data\, sea surface temperature prediction and flow over
a cylinder.\n
LOCATION:https://stable.researchseminars.org/talk/anumacs/12/
END:VEVENT
END:VCALENDAR