1,058 research outputs found
Centre for Mathematical Sciences Mathematical Statistics
No full textExplicit construction of GMRF approximations to generalised Matérn fields on irregular grids Finn Lindgren and H˚avard Ru
Modelling seismicity as a spatio-temporal point process using inlabru
Full text linkReliable deterministic prediction of earthquake occurrence is not possible at present, and may never be. In the absence of a reliable deterministic model, we need alternate strategies to manage the seismic hazard or the risk. This involves making statements of the likelihood or earthquake occurrence in space and time, including a fair and accurate description of the uncertainty around statements used in operational decision-making. Probabilistic Seismic Hazard Analysis (PSHA) and Operational Earthquake Forecasting (OEF) have the role of providing probabilistic statements on the hazard associated with earthquakes on long-term (decades to centuries) and short-term (days to decades) time frames respectively. Both
PSHA and OEF rely on a source model able to describe the occurrence of earthquakes.
PSHA models are commonly modelled using a spatially-variable Poisson process to describe earthquake occurrence. Therefore, they are calibrated on declustered catalogues which retains only the largest earthquakes in a sequence. OEF models, on the other hand, are commonly time-dependent models which describes the occurrence of all the events above a certain magnitude threshold including dependent events such as aftershocks or swarms. They are
calibrated on the full earthquake catalogue and provide accurate descriptions of the clustering process and the time-evolution of earthquake sequences. The Epidemic-Type Aftershock Sequence (ETAS) model is the most commonly used model as time-dependent seismicity model and belongs to the general class of Hawkes (or self-exciting) processes. Under the ETAS model, any earthquake in the sequence has the ability of inducing (or triggering) its own subsequence of earthquakes in a cascade of events, as commonly observed in nature. The earthquake catalogue is then the union of a set of events occurring independently from each other (background events) and a set of events which have been induced or triggered by
another (aftershocks).
The reliability of PSHA or OEF strategies depends upon the reliability of the source model used to describe earthquake occurrence. In order to improve the source model, we need the ability to (a) incorporate hypotheses on earthquake occurrence in a model, and (b) validate the model against observed data. Both tasks are problematic. Indeed, the complex mathematical form of the ETAS model requires ad-hoc methodologies to perform inference on the model parameters. These methodologies then need further modification if the classical
ETAS model is adjusted to introduce new hypotheses. Comparing forecasts produced by models incorporating different hypotheses which are and calibrated with different methods is problematic because it is difficult (if not impossible) to determine where the differences in the forecasts are coming from. Therefore, a unique framework capable of supporting ETAS models incorporating different hypotheses would be beneficial. Similarly, the validation step has to be done on models calibrated on the same data and producing forecasts for the same
spatio-temporal region. Moreover the validation must ultimately be done against future data, unknown in the moment in which the forecasts are produced, to ensure that no information about the data used to validate the models is incorporated in the models themselves. Hence, the Collaboratory for the Study of Earthquake Predictability (CSEP) has been founded with the role of gathering forecasting models and running fully-prospective forecasting experiments in an open environment. CSEP ensures that the models are validated fairly and using a set
of community-agreed metrics which measure the agreement between forecasts and data on the outcomes.
In this thesis, I present and apply a new Bayesian approximation technique for Hawkes process models (including ETAS). I also demonstrate the importance of one of the statistical properties that scores used to rank competing forecasts need to have in order to provide trustworthy results. The Bayesian framework allows an accurate description of the uncertainty around model parameters which can then be propagated to any quantity of interest. In
the context of Bayesian statistics, the most commonly used techniques to perform inference are Markov Chain Monte Carlo (MCMC) techniques which are sampling-based methods. Instead, I use the Integrated Nested Laplace Approximation (INLA) to provide a deterministic approximation of the parameter posterior distribution instead of the random sampling. INLA is faster than MCMC for problems involving a large number of correlated parameters and offers an alternative way to implement complex statistical models which are infeasible (from a computational point of view) with MCMC. This provides researchers and practitioners with a statistical framework to formulate ETAS models incorporating different hypotheses, produce forecasts that accounts for uncertainty, and test them using CSEP procedures. I build
on the work done to implement time-independent models for seismicity with INLA which provided a framework to study the effect of covariates such as depth, GPS displacement, heatflow, strain rate, and distance to the nearest fault but lacked the ability to describe the clustering process of earthquakes. I show that this work can be extended to include time-dependent Hawkes process models and run in a reasonable computational time using INLA. In this framework, the information from covariates can be incorporated both in modelling the rate of background events, and in modelling the number aftershocks. This resembles how information on covariates is incorporated in Generalized Linear Models (GLMs) which are widely used to study the effect of covariates on a range of phenomena. Indeed, this
work offers a way to borrow ideas and techniques used with GLMs and apply them to seismicity analyses. To make the proposed technique widely accessible, I have developed a new R-package called ETAS.inlabru which offers user-friendly access to the proposed methodology. The ETAS.inlabru package is based on the inlabru R-package which offers access to the INLA methodology. In this thesis, I compared our approach with the MCMC technique implemented through the bayesianETAS package and shows that ETAS.inlabru provides similar results to bayesianETAS, but it is faster, scales more efficiently increasing the amount of data, and can support a wider range of ETAS models, specifically those involving multiple covariates. I believe that this work provides users with a reliable Bayesian framework for the
ETAS model alleviating the burden of modifying/coding their own optimization routines and allowing more flexibility in the range of hypotheses that can be incorporated and validated. In this thesis, I have analysed the 2009 L’Aquila and 2016 Amatrice seismic sequences occurred in central Italy and found that the depth of the events have a negative effect on the aftershock productivity, and that models involving covariates show a better fit to the data than the classical ETAS model.
On the statistical properties that scores needs to posses to provide trustworthy rankings of competing forecasts, I focus on the notion of proper scores. I show that the Parimutuel Gambling (PG) score, used to rank forecasts in previous CSEP experiments, has been used in situations in which is not proper. Indeed, I demonstrate that the PG score is proper only in a specific situation and improper in general. I compare its performances with two proper alternatives: the Brier and the Logarithmic (Log) scores. The simulation procedure employed for this part of the thesis can be easily adapted to study the properties of other validation procedures as the ones used in CSEP or to determine important quantities for the experimental design such as the amount of data with which the comparison should be performed. This contributes to the wider discussion on the statistical properties of CSEP tests, and is an additional step in determining sanity-checks that scoring rules have to pass before being used to validate earthquake forecasts in CSEP experiments
Home in the Holy Land. A tale illustrating customs and incidents in modern Jerusalem By MRS Finn. London James Nisbet and Co., 21 Berners street. M.D.CCC. L.XVI
No full textDedication: by the author to the Countess of GriffordContent description: TitleIllustration: 4 (portraits ,varia ,)Pagination: PP8+520PVolumes: 1Text Genre:ProseIllustration: 4 (πορτραίτα ,άλλα θέματα ,
The future of pornography - panel debate. Speakers | Finn Mackay, Rowan Pelling, Peter Tatchell
No full textMany believe that porn's dark fantasies risk corrupting relationships and society. Has this arisen because pornography is largely created by men? Could feminist pornography featuring authentic sex, diverse bodies and female perspectives offer a truly liberating alternative? Or is porn fundamentally incompatible with intimacy and a problem for all of us until its abolished? Feminist thinker Finn Mackay, author of Belle de Jour: Diary of a London Call Girl Brooke Magnanti, human rights campaigner Peter Tatchell and Erotic Review editor Rowan Pelling imagine the future of pornography.In association with the New College of the Humanities
Author response: critical condition: replacing critical thinking with creativity by Patrick Finn
Full text linkFollowing Jodie Matthews’s review of his new book Critical Condition: Replacing Critical Thinking with Creativity, author Patrick Finn offers a response to Matthews’s reflections on his work. Finn outlines his approach to writing Critical Condition: a text that is designed to serve as an accessible ‘thought experiment’ that speaks across disciplines in order to explore its object, ‘critical thinking’. For Finn, critical thinking is a concept that appears integral to academic practice, yet remains under-defined and in need of further elucidation. Critical Condition therefore seeks to inspire collective reflection on critical thinking and its relationship with creativity
Agenda: Kjenn din Brecht, del 1
No full textVideoopptak av seminarAgenda starter høsten 2017 med et omfattende seminar om den tyske forfatteren og dramatikeren Bertolt Brecht (1898-1956) over to dager. Seminaret er et samarbeid med Morgenbladet.
Du vil få en innføring i Brechts liv og virke og presentasjoner av kunstnere som har forholdt seg til Brecht i praksis samt kritiske betraktninger rundt Brechts aktualitet i dag.
Brecht er en av det 20. århundres mest innflytelsesrike dramaturger og er en naturlig del av enhver kunstnerisk utdannelse. Han fikk stor betydning gjennom sine skuespill og dramaturgiske ideer. Til hans mest kjente stykker hører Galileis liv, Mor Courage, Det gode menneske fra Sezuan, Den kaukasiske kritringen og Tolvskillingsoperaen. Under nazistyret i Tyskland var Brecht i eksil i både Sverige og USA.
I 1948 flyttet han til Øst-Tyskland, hvor han grunnla og ledet det eksperimentelle teateret Berliner Ensemble, som fikk en banebrytende innflytelse på det europeiske etterkrigsteater. Han var overbevist kommunist, og har vært mye kritisert for sin uforbeholdne støtte til regimet.
Moderatorer: Lena Lindgren, politisk redaktør i Morgenbladet og Lotte Konow Lund, førsteamanuensis ved avdeling Kunst og håndverk, KHiO.
Program:
-Siren Leirvaag: "Brecht - en historisk figur eller relevant i dag?"
-Per Oskar Leu: "Om mørke tider"
-Therese Bjørneboe: "Brecht etter Brecht"
-Finn Iunker: "Passet er menneskets edleste del"
-Ragnhild Tronstad: Fridomens vegar som lærestykke
Using the SPDE approach for air quality mapping in Piemonte region
Full text linkIn this work we consider a geostatistical spatio-temporal model for PM10 concentration (particulate matter with an aerodynamic diameter of less than 10 μm) in the North-Italian region Piemonte. The model involves a Gaussian Field (GF) affected by a measurement error and a state process with a first order autoregressive dynamics and spatially correlated innovations. The main goal of this work is to propose an estimating and mapping strategy for such a model. This proposal is based on the work of Lindgren et al. (2011) that provides an explicit link between GFs and Gaussian Markov random fields (GMRF) through the Stochastic Partial Differential Equations (SPDE) approach. Thanks to the R library named INLA, the SPDE approach can be easily implemented providing results in reasonable computing time (with respect to other MCMC algorithms). For these reasons, the SPDE approach is proved to be a powerful strategy for modeling and mapping complex spatio-temporal phenomena
[CODE] gaussianboy/code-GFiSSA: 0.2
No full textCode for the GF-iSSA approach. The geoR package or perhaps the newest R version do not allow to run the simulation code. For this reason, we have uploaded all 9 simulation scenarios used in the paper consisting of 25 animal tracks (1000 locations) for each scenario.Rafael Arce Guillen, Finn Lindgren, Stefanie Muff, Thomas W. Glass, Greg A. Breed, & Ulrike E. Schlägel. (2023). gaussianboy/code-GFiSSA: 0.2 (0.2). Zenodo. https://doi.org/10.5281/zenodo.837958
Bayesian Spatial Modelling with R-INLA
Full text linkThe principles behind the interface to continuous domain spatial models in the R- INLA software package for R are described. The integrated nested Laplace approximation (INLA) approach proposed by Rue, Martino, and Chopin (2009) is a computationally effective alternative to MCMC for Bayesian inference. INLA is designed for latent Gaussian models, a very wide and flexible class of models ranging from (generalized) linear mixed to spatial and spatio-temporal models. Combined with the stochastic partial differential equation approach (SPDE, Lindgren, Rue, and Lindström 2011), one can accommodate all kinds of geographically referenced data, including areal and geostatistical ones, as well as spatial point process data. The implementation interface covers stationary spatial mod- els, non-stationary spatial models, and also spatio-temporal models, and is applicable in epidemiology, ecology, environmental risk assessment, as well as general geostatistics
Bayesian Spatial Modelling with R-INLA
Full text linkThe principles behind the interface to continuous domain spatial models in the R- INLA software package for R are described. The integrated nested Laplace approximation (INLA) approach proposed by Rue, Martino, and Chopin (2009) is a computationally effective alternative to MCMC for Bayesian inference. INLA is designed for latent Gaussian models, a very wide and flexible class of models ranging from (generalized) linear mixed to spatial and spatio-temporal models. Combined with the stochastic partial differential equation approach (SPDE, Lindgren, Rue, and Lindström 2011), one can accommodate all kinds of geographically referenced data, including areal and geostatistical ones, as well as spatial point process data. The implementation interface covers stationary spatial mod- els, non-stationary spatial models, and also spatio-temporal models, and is applicable in epidemiology, ecology, environmental risk assessment, as well as general geostatistics
- …
