Past events

TUESDAY Statistics Seminar: Aliaksandr Hubin

Tuesday, April 25, 2017 - 14:15

Speaker: Aliaksandr Hubin (Department of Mathematics, University of Oslo) 

Title: A novel algorithmic approach to Bayesian Logic Regression

Abstract: Logic regression was developed more than a decade ago as a tool to construct predictors from Boolean combinations of binary covariates. It has been mainly used to model epistatic effects in genetic association studies, which is very appealing due to the intuitive interpretation of logic expressions to describe the interaction between genetic variations. We introduce an advanced evolutionary algorithm called GMJMCMC (Genetically modified Mode Jumping Markov Chain Monte Carlo) to perform Bayesian model selection in the space of logic regression models. After describing the algorithmic details of GMJMCMC we perform a comprehensive simulation study that illustrates its high performance for logic terms of various complexity. Specifically GMJMCMC is shown to be able to identify three-way and even four-way interactions with large power, whilst keeping FDR reasonably small. This level of complexity has not been achieved by previous implementations of Bayesian logic regression. Finally, we apply GMJMCMC to reanalyze QTL mapping data for Recombinant Inbred Lines in Arabidopsis thaliana and from a backcross population in Drosophila where we identify several interesting epistatic effects.

Location: 
Department of Mathematics, room Sverdrups plass (lunch area) on the 8th floor of Niels Henrik Abels hus

THURSDAY Biostatistical Seminar: Georg Heinze

Thursday, March 30, 2017 - 14:15

Speaker: Georg Heinze, Professor, Center for Medical Statistics, Informatics, and Intelligent Systems (Section for Clinical Biometrics), Medical University of Vienna, Austria

Title: The multiple faces of shrinkage

Coffee/tea served from 14:15 and the talk will start around 14:30.

Abstract: Shrinkage is a result of overfitting, if regression models are estimated with small or sparse data sets.  In such situations predictions for new subjects are often ‘too extreme’ and their real outcomes are closer to an overall mean, i.e. they appear to be ‘shrunken’. Interestingly, ‘shrinkage’ is also often used to denote estimators that aim at anticipating shrinkage effects and preventing its occurrence. This duality has often caused confusion.

Shrinkage estimators can serve various purposes. Some methods were developed to optimize the calibration of prediction models. Other methods should reduce bias away from zero, which in logistic regression problems with small samples can be severe, but is absent, e.g., in linear regression. Another purpose could be to improve the accuracy, i.e., to reduce mean squared error of predictions or of effect estimates. Irrespective of their purpose, some of these methods can have a Bayesian motivation, where prior belief about possible values of common estimands such as log odds ratios is expressed as prior distributions centered at zero.

Shrinkage estimators can be constructed by maximizing a likelihood function penalized by an additional function of the parameters, which pulls estimates towards zero. Ridge and lasso regression are well-known examples, and so is Firth’s penalized likelihood. Other shrinkage estimators are constructed differently, e.g., estimating and applying post-estimation shrinkage factors by resampling methods. It is less well known that also classical variable selection methods can be interpreted as shrinkage estimators.

The talk will mainly focus on the setting of logistic regression with rare events. After a general introduction, we will compare shrinkage estimators by their assumed ‘pessimism’, i.e., the amount of overfitting that they anticipate (Kammer et al, 2017). Subsequently, we will investigate the improvement in accuracy of parameter estimates and predicted probabilities implicated with various shrinkage methods (Puhr et al, 2017). Finally, we will briefly discuss Bayesian noncollapsibility, i.e., likelihood penalization resulting in undesired anti-shrinkage, which can affect all well-known shrinkage estimators (Greenland, 2010; Geroldinger et al, 2017).

Organizer: Oslo Centre for Biostatistics and Epidemiology (OCBE), Research group in Statistics and Biostatistics, Dept. of Mathematics, UiO and Big Insight

Location: 
Domus Medica, room 2240 (Dep. of Biochemistry)

Wednesday Lunch: Steffen Sjursen

Wednesday, March 29, 2017 - 12:00

Speaker: Steffen Sjursen, Senior analyst, group risk modelling, DNB. PhD UiO: Stochastic Optimal Control and time changed Lévy noises.

Title: Calculating probability of default when available data varies greatly between different customers.

Abstract: TBA

The lunch starts at 12:00, and the talk will start around 12:20.

NB: Wednesday BigInsight Lunches are open to staff and students from any of the BigInsight partners, including UiO, but not to others

 

Location: 
Spiseriet, Norwegian Computing Center

TUESDAY Statistics Seminar: Nicola Lunardon

Tuesday, March 28, 2017 - 14:15

Speaker: Nicola Lunardon, University of Milano-Bicocca

Title: On bias prevention and incidental parameters.

Abstract: Firth (1993, Biometrika) introduced the bias prevention approach with the aim to reduce the bias of the maximum likelihood estimator. It is shown that the methodology is also effective in reducing the sensitivity of the derived inferential procedures in models involving incidental parameters.

Location: 
Department of Mathematics, room Sverdrups plass (lunch area) on the 8th floor of Niels Henrik Abels hus

THURSDAY Biostatistical Seminar: Harald Binder

Thursday, March 16, 2017 - 14:15

Speaker: Harald Binder, Professor, Institute of Medical Biostatistics, Epidemiology and Informatics, University Medical Center, Johannes Gutenberg University Mainz, Germany.

Title: Two uses of stagewise regression: from landmarking in cancer patients to deep learning for SNPs

Coffee/tea served from 14:15 and the talk will start around 14:30.

Abstract: Regularized techniques allow to estimate the parameters of a regression model and to perform variable selection even when the number of covariates is larger than the number of observations.  Penalized likelihood approaches are the most prominent way for performing such regularization, e.g. the Lasso for variable selection. Stagewise regression techniques, also known as componentwise boosting, provide an alternative to the latter. Starting from estimates all equal to zero, one parameter is updated in each of a potentially large number of steps. Typically, this does not stop at the maximum of a global criterion, such as a penalized log-likelihood, which makes such approaches difficult to treat analytically might be seen as a disadvantage. Yet, there are two distinct advantages: optimization in settings where no (overall) likelihood is a available, and speed. I will show two applications, where these advantages play out. In a dynamic prediction application with data from hepatocellular carcinoma patients, effects of patient characteristics on the cumulative incidence of death (in presence of competing risks) are investigated using pseudo-value regression models at different landmarks. Stagewise regression allows to couple variable selection between landmarks without enforcing similarity between parameter estimates. As a result the approach selects a set of variables that is relevant for all landmarks. In a second application, the aim is to search for potentially complex patterns in molecular data, namely single nucleotide polymorphisms (SNPs), which might be relevant for patient prognosis. The modeling of patterns is performed by a deep learning approach, specifically deep Boltzmann machines, which cannot be applied in a straightforward manner if the number of SNPs is larger than the number of observations. Thus, stagewise regression is used to obtain a rather crude (implicit) estimate of the joint distribution of SNPs, which forms the basis for partitioned training of deep Boltzmann machines. The resulting SNP patterns are seen to be relevant with respect to clinical outcomes.

Organizer: Oslo Centre for Biostatistics and Epidemiology (OCBE), Research group in Statistics and Biostatistics, Dept. of Mathematics, UiO and Big Insight

Location: 
Domus Medica, room 2240 (Dep. of Biochemistry)

Wednesday Lunch: Einar Broch Johnsen

Wednesday, March 15, 2017 - 12:00

Speaker: Einar Broch Johnsen, professor at the Department of Informatics, University of Oslo.

Title: Cloud computing: Predicting behavior using models.

Abstract: We discuss cloud computing and how semantic models can be used to predict the behavior of software running on the cloud.
Software on the cloud differs from regular software in that it can adapt its configuration elastically depending on its computational load.
The talk will be based on research from the EU FP7 project Envisage, coordinated by Johnsen 2014–2016.

The lunch starts at 12:00, and the talk will start around 12:20.

Location: 
Sverdrups plass (lunch area), 8th floor N. H. Abels hus, Department of Mathematics

THURSDAY Biostatistical Seminar: Hein Putter

Thursday, March 9, 2017 - 14:15

Speaker: Hein Putter, Proffesor, Department of Medical Statistics and Bioinformatics, Leiden University Medical Center, The Netherlands

Title: Non-parametric estimation of transition probabilities in non-Markov multi-state models: the landmark Aalen-Johansen estimator

Coffee/tea served from 14:15 and the talk will start around 14:30.

Abstract: The topic non-parametric estimation of transition probabilities in non-Markov multi-state models has seen a remarkable surge of activity recently. Two recent papers have used the idea of subsampling in this context. The first paper, by de Uña Alvarez and Meira-Machado, uses a procedure based on (differences between) Kaplan–Meier estimators derived from a subset of the data consisting of all subjects observed to be in the given state at the given time. The second, by Titman, derived estimators of transition probabilities that are consistent in general non-Markov multi-state models. Here, we show that the same idea of subsampling, used in both these papers, combined with the Aalen–Johansen estimate of the state occupation probabilities derived from that subset, can also be used to obtain a relatively simple and intuitive procedure which we term landmark Aalen–Johansen. We show that the landmark Aalen–Johansen estimator yields a consistent estimator of the transition probabilities in general non-Markov multi-state models under the same conditions as needed for consistency of the Aalen–Johansen estimator of the state occupation probabilities. Simulation studies show that the landmark Aalen–Johansen estimator has good small sample properties and is slightly more efficient than the other estimators.

Organizer: Oslo Centre for Biostatistics and Epidemiology (OCBE), Research group in Statistics and Biostatistics, Dept. of Mathematics, UiO and Big Insight

Location: 
Domus Medica, room 2240 (Dep. of Biochemistry)

TUESDAY Statistics Seminar: Lan Zhang

Tuesday, March 7, 2017 - 14:15

Speaker: Lan Zhang​, University of Illinois at Chicago

Title: The S-TSRV: Robust High Frequency Estimation

Abstract: In this paper, we derive a new algebraic property of two scales estimation in high frequency data, under which the effect of sampling times is cancelled to high order. This is a particular robustness property of the two scales construction. In general, irregular times can cause problems in estimators based on equidistant observation of (trade or quote) times.

The new algebraic property can be combined with pre-averaging, giving rise to the smoothed two-scales realized volatility (S-TSRV). We derive a finite sample solution to controlling edge effects and for handling endogenous observation times and asynchronously observed multivariate data. In connection with this development, we use the algebraic approach to define a version of the S-TSRV which has particularly small edge effect in microstructure noise. The main result of the paper is a representation of the statistical error of the estimator in terms of simple components. As an application, the paper develops a central limit theory for multivariate volatility estimators.

Location: 
Department of Mathematics, room Sverdrups plass (lunch area) on the 8th floor of Niels Henrik Abels hus

Wednesday Lunch: Karianne de Bruin

Wednesday, March 1, 2017 - 12:00

Speaker: Karianne de Bruin, senior researcher at Cicero, and researcher at the CALM team at Wageningen Environmental Research, Netherlands.

Title: Adaption decision-making under uncertainty

Abstract: The presentation will highlight the theoretical background of decision-making under uncertainty in the context of adapting to climate change and will discuss how this fits into the context of 'climate services', services which aim to facilitate the production, translation and tailoring of climate information to build long term resilience.

The lunch starts at 12:00, and the talk will start around 12:20.

Location: 
Spiseriet, Norwegian Computing Center

TUESDAY Statistics Seminar: Riccardo De Bin

Tuesday, February 28, 2017 - 14:15

Speaker: Riccardo De Bin (Department of Mathematics,UiO)

Title: Strategies to handle mandatory covariates using model- and likelihood-based boosting

Abstract:  Among the iterative methods exploited during recent years in statistical practice, particular attention has been focused on boosting. Originally developed in the machine learning community to handle classification problems, boosting has been successfully translated into the statistical field and extended to many statistical problems, including regression and survival analysis. In a parametric framework, the basic idea of boosting is to provide estimates of the parameters by updating their values iteratively: at each step, a weak estimator is fitted on a modified version of the data, with the goal of minimizing a loss function. Thanks to its resistance to overfitting, boosting is particularly useful in the construction of prediction models. Its iterative nature, moreover, allows straightforward adaptations to cope with high-dimensional data. In this talk, we first review and contrast two well-known boosting techniques, model-based boosting and likelihood-based boosting. We note that in the simple linear regression case they lead to the same results, provided there is a specific choice for their tuning parameters. This is not the case for more complex situations. As an example, we show the differences in survival analysis under the proportional hazards assumption. As a main contribution of the talk, we analyze strategies to include mandatory variables, i.e. those variables that for some reasons must enter in the final model, in a statistical model using the two boosting techniques. In particular, we examine solutions currently only considered for one and explore the possibility of extending them to the other. We show the importance of a good handling of mandatory variables in a real data example.

Location: 
Department of Mathematics, room Sverdrups plass (lunch area) on the 8th floor of Niels Henrik Abels hus

Wednesday Lunch: Privacy and Big Data

Wednesday, February 15, 2017 - 12:00

Speaker: Catharina Nes, Specialist director, research and analytics at the Norwegian Data Protection Authority (Datatilsynet).

The lunch starts at 12:00, and the talk will start around 12:20.

Location: 
Sverdrups plass (lunch area), 8th floor N. H. Abels hus, Department of Mathematics

Wednesday Lunch: A gentle inroduction to Bayesian Nonparametrics

Wednesday, February 1, 2017 - 12:00

Speaker: Nils Lid Hjort, Professor at UiO, Dept of Mathematics

Abstract: Bayesian Nonparametrics is about prior distributions on big and complicated spaces, such as the set of all continuous densities or regression functions, and then working out the mathematics to characterise the posterior distributions, along with clever computational or simulation schemes for inferences, predictions, classifications. Such methods are finding their ways also into Machine Learning applications. I will provide a necessarily short and hopefully gentle introduction to the field.

The lunch starts at 12:00, and the talk will start around 12:20.

Location: 
Spiseriet, Norwegian Computing Center

TUESDAY Statistics Seminar: Inge S. Helland

Tuesday, January 31, 2017 - 14:15

Speaker: Inge S. Helland (Professor emeritus at Department of Mathematics, UiO)

Title: Symmetry and model reduction

Abstract: Statistics is the basis for most empirical sciences, and an interesting question is whether one can find links to other, complementary scientific cultures by taking statistical theory as a point of departure. I will show that the answer is yes for at least two cases if one adds the following structure to the statistical model paradigm: By suitable symmetry assumptions there may be a group of transformations defined on the sample space and a corresponding group of transformations defined on the parameter space. If the parameter group is not transitive, it induces several orbits on the parameter space. I will postulate that any model reduction should be to an orbit or to a set of orbits of the chosen group. First I illustrate this rule by giving several statistical examples. Then I show how the rule leads to the partial least squares model, and I indicate how one can derive the quantum theory for electron spin in this way. Some recent results on quantum probability are mentioned.

Location: 
Department of Mathematics, room Sverdrups plass (lunch area) on the 8th floor of Niels Henrik Abels hus

TUESDAY STATISTICS SEMINAR: Daniel Roy

Tuesday, December 20, 2016 - 14:15

Speaker: Daniel Roy (Department of Statistical Sciences, University of Toronto)

Title: Extended admissible if and only if nonstandard Bayes

Abstract:  For finite parameter spaces under finite loss, every Bayesian procedure derived from a prior with full support is admissible, and every admissible procedure is Bayes. This relationship begins to break down as we move to continuous parameter spaces. Under some regularity conditions, admissible procedures can be shown to be the limits of Bayesian procedures. Under additional regularity, they are generalized Bayesian, i.e., they minimize the Bayes risk with respect to an improper prior. In both these cases, one must venture beyond the strict confines of Bayesian analysis. Using methods from mathematical logic and nonstandard analysis, we introduce the class of nonstandard Bayesian decision procedures---namely, those whose Bayes risk with respect to some prior is within an infinitesimal of the optimal Bayes risk.  Without any regularity conditions, we show that a decision procedure is extended admissible if and only if its nonstandard extension is nonstandard Bayes. We apply the nonstandard theory to derive a purely standard theorem: on a compact parameter space, every extended admissible estimator is Bayes if the risk function is continuous.
Joint work with Haosui Duanmu.​

This will be the last TUESDAY STATISTICS SEMINAR of 2016. We look forward to see you all again in 2017!

Location: 
Department of Mathematics, room Sverdrups plass (lunch area) on the 8th floor of Niels Henrik Abels hus

Wednesday Lunch: Forecasting power systems

Wednesday, November 30, 2016 - 12:00

Speaker: Alex Lenkoski, Senior Research Scientist at the Norwegian Computing Center

The lunch starts at 12:00, and the talk will start around 12:20.

Location: 
Spiseriet, Norwegian Computing Center