Seminar on Stochastic Models

When

Who

Topic

Latent Markov models for the analysis of the Russia Longitudinal Monitoring Survey

The work describes an application of a class of models for longitudinal data, called Latent Markov models, to the Russia Longitudinal Monitoring Survey. Latent Markov (LM) models are an important class of latent variable models aimed for the study of this type of data. The dataset chosen to be analysed by applying the LM models is the Russia Longitudinal Monitoring Survey, which is a panel survey, obtained from face-to-face interviews and surveyed annually since 1992. The aim of this work is to detect different groups of individuals having different levels of satisfaction and how these degrees of satisfaction evolve across years, depending on the socio-demographic and economical individual covariates.

I will shortly show you the possibilities of Shiny package in R (a package that enables you to make interactive dashboards very easily). I will show how to make you Shiny applications available on shiny.ut.ee server or shinyapps.io server.

Some examples of Shiny dashboards:
- statistika.ut.ee
- a students' Shiny app on videogame data (see more on course "Andmeteadus ja visualiseerimine"  - Caution! Not all might work. More on this in the seminar).

Tõnu Kollo will give an overview of the paper written jointly with Marju Valge about testing basic covariance structures for the t-distributed population.

About European Day of Statistics and its celebration in Paris last week and an overview about Holger Rootzen's talk about extreme value distributions.

• Quantitative methods for implementing TF strategies
• The convexity profile of TF and other systematic startegies
• New portfolio theory incorporating tail risk

Ruin probability for merged risk processes with correlated arrivals

Kaur Alasoo (LTAT)

Ravel Riik (MS magistrant)

“Quantifying variance explained by gene-environment interactions in molecular data"/ "Geeni-keskkonna koosmõjude poolt kirjeldatud varieeruvuse hindamine molekulaarsetes andmetes"

"Kodulaenude portfelli riskide korrigeermisest Kalmani filtri abil"

Visit to the University of Utrecht. Joint interests. Surprises in estimation under nonresponse

Genetic risk score for breast cancer

16.10.18

Mohammad Jamsher Ali

Wider Bootstrap Confidence Interval. The estimation of the unknown parameter is the typical problem in applied statistics. It is obvious to raise the questions that 

* what estimator should be used and

* how accurate is the chosen estimator as an estimation of our parameter

The bootstrap is the general methodology to examine the accuracy of our estimator. It is a computer-based method first introduced by B. Efron in 1979. 

According to Efron, using skewness of the estimator it is possible to construct bootstrap confidence interval which is wider than standard confidence interval.

25.09.18

Maarika Traat

Pulsatility of small cortical veins using phase contrast magnetic resonance imaging at 7 Tesla.

Venous pulsatility measured from large cerebral veins has been shown to change in pathologies such as normal pressure hydrocephalus, Alzheimer’s disease, vascular dementia, mild cognitive impairment and multiple sclerosis. Venous pulsatility also increases with age. Recent improvements in the resolution of imaging techniques have made it feasible to explore smaller vessels. In my talk I give an overview of my Master’s project at the brain imaging centre of Cardiff University where, for the first time, we demonstrated the pulsatility of small cortical veins using phase-contrast magnetic resonance imaging at 7 tesla. Another very important finding was the statistically significant lag detected between the arrival of the cardiac pulse wave into the small cortical arteries and the small cortical veins. Measurements from smaller veins enhance the model of cerebral haemodynamics and add valuable insight into the medical conditions where intracranial pressure or pulsatility has been compromised, as well as into normal aging.

18.09.18 

Konverentsidest, millel osaletud

04.09.18

Annika Krutto

valmiva doktoritöö "Empirical Cumulant Function Based Estimation in Stable Laws" tutvustamine. Juhendaja Tõnu Kollo.

28.11.17

Joonas Sova

07.11.17

Kaur Lumiste

valmiva doktoritöö Improving accuracy of survey estimators by using auxiliary information in data collection and estimation stages tutvustamine. Juhendaja Imbi Traat.

06.11.17

Paul Tammo

valmiva doktoritöö Closed maximal regular one-sided ideals in topological algebras tutvustamine. Juhendajad Mart Abel ja Mati Abel.

24.10.17

Annika Krutto

Stabiilse jaotuse parameetrite hindamine kumulantfunktsioonist. Stabiilsed jaotused moodustavad laia 4-parameetrilise tõenäosusjaotuste klassi. See eristati 1920ndatel kui sõltumatute sama jaotusega juhuslike suuruste normeeritud summa piirjaotus. Klass hõlmab nii kergete/raskete sabadega kui sümmeetrilised/asümmeetrilised juhud ja on leidnud ulatuslikku rakendust erinevates valdkondades. Stabiilse jaotuse parameetrite hindamise muudab keerukaks analüütilise jaotusfunktsiooni puudumine (välja arvatud erijuhud nagu  normaal-, Cauchy ja Levy jaotus). Viimastel aastakümnetel on  esitatud mitmeid lahendusi kuid neil kõigil on omad kitsendused. Käesoleva uurimus eesmärk on edasi arendada 1970ndatel välja pakutud üldistatud momentide meetodit.

17.10.2017

Meelis Kull (LTAT)

Masinõppe algoritmide liigne enesekindlus ja kuidas seda vältida. Keerulised intelligentsed süsteemid nagu isesõitvad autod ja meditsiinilised ekspertsüsteemid rakendavad masinõppega ehitatud klassifikaatoreid. Tihti on tarvis, et need klassifikaatorid väljastaks koos ennustusega ka enesekindluse määra, sest see võimaldab süsteemil madala enesekindluse korral teha turvalisemaid valikuid. Seepärast on väga oluline, et klassifikaator ei oleks liiga enesekindel, sest see suurendaks väga kalliste vigade riski. Paraku on masinõppe meetoditega saadavad hinnangud siiski tihti liiga enesekindlad. Seminaris räägin millest see põhjustatud on ning kuidas kalibreerimise abil liigsest enesekindlusest lahti saada.