BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Bayesian MANOVA for the combined evaluation of handwriting evidenc
 e
DTSTART;VALUE=DATE-TIME:20230628T132000Z
DTEND;VALUE=DATE-TIME:20230628T134000Z
DTSTAMP;VALUE=DATE-TIME:20261018T011253Z
UID:indico-contribution-200-1018@cern.ch
DESCRIPTION:Speakers: Lampis Tzai ()\nForensic science is a broad field th
 at uses scientific principles and technical methods to help with the evalu
 ation of evidence in legal proceedings of criminal\, civil\, or administra
 tive nature. Forensic scientists examine recovered traces that can be give
 n by glass fragments\, fingerprints\, body fluids\, textile fibers\,\ndigi
 tal device data and handwriting. The handwriting examination is a well-kno
 wn field of analysis. Consider a case involving a handwritten document of 
 unknown origin. Handwritten features extracted from this questioned docume
 nt will be compared to those extracted from a document written by a person
  that is suspected of being at the origin of the anonymous document. The p
 ropositions of interest are the following:\n\n$H_p$: the suspect is the au
 thor of the manuscript\;\n\n$H_d$: the suspect is not the author of the ma
 nuscript.\n\nHandwriting individualization is still largely dependent on t
 he experience of the document examiner\, though studies have been conducte
 d with the aim of supporting handwriting examiners to reduce the degree of
  subjectivity of their expertise. Marquis et al. (2005) proposed to increa
 se the degree of objectivity of handwriting analyses by implementing eleme
 nts of Fourier analysis in order to describe the contour shape of loops of
  characters. Specifically\, the characters that contain loops can be descr
 ibed by means of Fourier descriptors\, which can be used to characterize t
 he shape complexity and other geometric attributes. The\nanalyses conducte
 d showed that these features have a good discriminating power.\nWith the a
 im of implementing the use of these handwriting features for handwriting i
 dentification\, Bozza et al. (2008) proposed a Bayesian probabilistic appr
 oach by modeling the data with multivariate Normalinverse-Wishart distribu
 tion (NIW). The value of the evidence is subsequently assessed by means of
  the Bayes factor\, which can be interpreted as a measure of the strength 
 of support provided by the evidence in favor of the hypothesis $H_p$ again
 st the hypothesis $H_d$. This approach was accomplished to take into accou
 nt the correlation between variables\, the variability between-writers and
  within-writer variability. However\, the above model is implemented separ
 ately for each different type of handwritten character. This can be proble
 matic because it can lead to a different conclusion depending on the type 
 of character that is retained.\n\nIn this research\, it is proposed the im
 plementation of a Bayesian Multivariate Analysis of Variance (MANOVA) via 
 using the loop characters as predictors. The indicator of the loop charact
 er is transformed into a dummy variable (corner-point representation)\, so
  that it is possible to model variables describing\nthe handwriting charac
 ters jointly taking into account the variability between characters\, the 
 variability between-writers for every character and within-writer variabil
 ity. The Bayesian MANOVA is compared with the two-level random effect mode
 l (NIW) proposed by Bozza et al. (2008)\, that is implemented by modelling
 \nall characters jointly or separately. Three different methods for estima
 ting the marginal likelihoods are used\; the Generalized Harmonic Mean\, t
 he Laplace-Metropolis and the Bridge Sampling. Finally\, the performances 
 of the NIW and MANOVA models are compared with those of an alternative one
 \, where a conjugate approach is chosen. This does not allow to model the 
 within and between variation separately\, but the marginals can be obtaine
 d analytically.\n\nFirstly\, we estimate the Bayes factor of the two data 
 models for each writer to determine which model is more compatible with th
 e data. Secondly\, there have been selected handwriting features originati
 ng from the same writer or from different writers to evaluate the rate of 
 false negatives (that is cases where the BF is smaller than one for charac
 ters originating from the same source) and false positives (that is cases 
 where the BF is greater than one for characters originating from different
  sources). Finally\, the sensitivity of models is examined in two critical
  aspects: the misleading background information and the choice of degrees 
 of freedom for the Wishart-inverse distribution that is used to model the 
 handwriting variability. With reference to the misleading background infor
 mation\, the prior distributions were elicited by selecting writers charac
 terized by either small or marked differences.\n\n**Keywords** : 1. Handwr
 iting Evidence 2. Fourier Analysis 3. Multivariate Bayesian Modelling 4. B
 ayes Factor 5. Sensitivity\n\n**References**\n\nBozza\, S.\, Taroni\, F.\,
  Marquis\, R. & Schmittbuhl\, M. (2008)\, ‘Probabilistic evaluation of h
 andwriting evidence:\nlikelihood ratio for authorship’\, Journal of the 
 Royal Statistical Society: Series C (Applied Statistics)\n57(3)\, 329–34
 1.\n\nMarquis\, R.\, Schmittbuhl\, M.\, Mazzella\, W. D. & Taroni\, F. (20
 05)\, ‘Quantification of the shape of handwritten\ncharacters: a step to
  objective discrimination between writers based on the study of the capita
 l character o’\,\nForensic Science International 150(1)\, 23–32.\n\nht
 tps://indico.unina.it/event/67/contributions/1018/
LOCATION:Department of Political Sciences Aula Spinelli
URL:https://indico.unina.it/event/67/contributions/1018/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Variable selection via ranking in generalized linear models
DTSTART;VALUE=DATE-TIME:20230628T140000Z
DTEND;VALUE=DATE-TIME:20230628T142000Z
DTSTAMP;VALUE=DATE-TIME:20261018T011253Z
UID:indico-contribution-200-1007@cern.ch
DESCRIPTION:Speakers: Marcella Niglio (Università degli Studi di Salerno)
 \nIn many empirical domains\, the availability of ultrahigh-dimensional da
 ta has led to the development of feature screening and variable selection 
 procedures aiming to detect the informative variables of datasets and cons
 equently remove unimportant features.\nIn this context\, we propose a rank
 ing-based variable selection procedure that extends the Ranking Based Vari
 able Selection technique (Baranowski et al.\, 2020) to general linear regr
 ession models.\nWe explore the performance of our proposal using both simu
 lated and empirical data. The algorithm is compared to two competitors: i)
  the Extended BIC (Chen and Chen\, 2012)\; ii) the variable selection proc
 edure based on the combination of the Sure Independence Screening (Fan and
  Song\, 2010) and the Elastic Net (Zou and Hastie\, 2005).\n\n**References
 **\n\nBaranowski R\, Chen Y\, Fryzlewicz P (2020)\, Ranking-based variable
  selection for high-dimensional data\, Statistica Sinica\, 30(3)\, 1485-15
 16.\n\nChen J\, Chen Z (2012)\, Extended BIC for small-n-large-P sparse GL
 M. Statistica Sinica\, 22(2)\, 555-574.\n\nFan J\, Song R (2010)\, Sure in
 dependence screening in generalized linear models with NP-dimensionality. 
 The Annals of Statistics\, 38(6)\, 3567-3604.\n\nZou H\; Hastie T (2005)\,
  Regularization and Variable Selection via the Elastic Net. Journal of the
  Royal Statistical Society\, Series B\, 67(2)\, 301-320.\n\nhttps://indico
 .unina.it/event/67/contributions/1007/
LOCATION:Department of Political Sciences Aula Spinelli
URL:https://indico.unina.it/event/67/contributions/1007/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Objective Shrinkage Priors Via Imaginary Data
DTSTART;VALUE=DATE-TIME:20230628T134000Z
DTEND;VALUE=DATE-TIME:20230628T140000Z
DTSTAMP;VALUE=DATE-TIME:20261018T011253Z
UID:indico-contribution-200-1001@cern.ch
DESCRIPTION:Speakers: Dimitris Fouskakis (National Technical University of
  Athens)\nIn this work\, focus is given in the Bayesian variable selection
  problem for high-dimensional linear regression problems. The use of shrin
 kage priors\, when the number n of available observations is less than the
  number p of explanatory variables\, is a well-established method\, which 
 shares great theoretical and empirical properties. By using imaginary data
  and shrinkage priors as baseline priors\, under the Power-Expected-Poster
 ior (PEP) prior methodology\, objective shrinkage priors are being created
 . In addition\, we explore the idea of augmenting the imaginary design mat
 rix in order to make it with orthogonal columns and thus to produce indepe
 ndent PEP-shrinkage priors\, based on default baseline priors. Under this 
 setup\, properly chosen hyperpriors are placed on the power parameters of 
 the PEP methodology\, in order to produced mixtures of independent priors 
 suitable for the variable selection problem when n << p. This second appro
 ach provides us with algorithmically flexibility and less time-consuming p
 rocedures. We check the theoretical properties of our proposed methods and
  we explore their behavior via simulated studies.\n\nKeywords: Bayesian Va
 riable Selection\; Imaginary Data\; Objective Priors\; Shrinkage Priors\n\
 nhttps://indico.unina.it/event/67/contributions/1001/
LOCATION:Department of Political Sciences Aula Spinelli
URL:https://indico.unina.it/event/67/contributions/1001/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Subdata selection for big data regression based on leverage scores
DTSTART;VALUE=DATE-TIME:20230628T130000Z
DTEND;VALUE=DATE-TIME:20230628T132000Z
DTSTAMP;VALUE=DATE-TIME:20261018T011253Z
UID:indico-contribution-200-1004@cern.ch
DESCRIPTION:Speakers: Vasilis Chasiotis (Athens University of Economics an
 d Business\, Department of Statistics)\nData continues to become more abun
 dant\, and so the datasets that contain it. Even though big datasets can p
 resent insights and opportunities\, they can pose significant challenges w
 hen it comes to statistical analysis. One of the biggest challenges\, requ
 ired to process and analyze large datasets\, is the computational resource
 s. Regression can be problematic in case of big datasets\, due to the huge
  volumes of data. A standard approach is subsampling that aims at obtainin
 g the most informative portion of the big data. We consider an approach ba
 sed on leverages scores\,  already existing in the current literature for 
 the selection of subdata for linear model discrimination. However\, we hig
 hlight its importance on the selection of data points that are the most in
 formative for estimating unknown parameters. We conclude that the approach
  based on leverage scores improves existing approaches\, providing simulat
 ion experiments as well as a real data application.\n\nhttps://indico.unin
 a.it/event/67/contributions/1004/
LOCATION:Department of Political Sciences Aula Spinelli
URL:https://indico.unina.it/event/67/contributions/1004/
END:VEVENT
END:VCALENDAR