Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca

Page created by Claude Salinas
 
CONTINUE READING
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Presentazione ai dottorandi: The Cattolica reseach group

                  Raffaele Argiento
                raffaele.argiento@unicatt.it

                         Bicocca
                Milano, 24 Settembre 2020

                                               R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Presentazione ai dottorandi: The Cattolica reseach group

                  Raffaele Argiento
                raffaele.argiento@unicatt.it

                         Bicocca
                Milano, 24 Settembre 2020

                                               R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Biostatistics

Collaborations: Rice University & MD Anderson, TX USA – Na-
tional Univerisy of Singapore (NUS) – College of Public Healt,
National Taiwan University.

                                    R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Biostatistics

                                   Collaborations: Rice University & MD Anderson, TX USA – Na-
                                   tional Univerisy of Singapore (NUS) – College of Public Healt,
                                   National Taiwan University.

Microbiome data: modelling the association
between bacteria taxa e nutrients

                                                                       R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Biostatistics

                                   Collaborations: Rice University & MD Anderson, TX USA – Na-
                                   tional Univerisy of Singapore (NUS) – College of Public Healt,
                                   National Taiwan University.

Microbiome data: modelling the association        Population structure:      Genetic Diversity
between bacteria taxa e nutrients                 (subpopulations)

                                                                          R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Biostatistics Data Privacy

                               Collaborations: University of Firenze – University of Modena e
                               Reggio Emilia – National Technical University of Athens.

Protein data: Discover relationship between
proteins, causal inference

                                                                   R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Data Privacy

                               Collaborations: Università della Svizzera Italiana – Axa –

Protein data: Discover relationship between      Fraud detection: shorten the delay from the
proteins, causal inference                       occurrence of the fraud to its detection;

                                                                      R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Sport Data & Healthcare

Collaborations: Kent University (UK) – Università di Torino – Po-
litecnico Milano – Victoria University, Melbourne (AUS)

                                                                    R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Sport Data & Healthcare

 Collaborations: Kent University (UK) – Università di Torino – Po-
 litecnico Milano – Victoria University, Melbourne (AUS)

Sport analytics: performances predictions,
doping detection

                                                                     R. Argiento   Milano, 24 Settembre
Raffaele Argiento Presentazione ai dottorandi: The Cattolica reseach group - Bicocca
Applications: Sport Data & Healthcare

 Collaborations: Kent University (UK) – Università di Torino – Po-
 litecnico Milano – Victoria University, Melbourne (AUS)

Sport analytics: performances predictions,         Avis data: Clustering donors for customized
doping detection                                   advertising

                                                                         R. Argiento   Milano, 24 Settembre
Applications: Atmospheric Environment

• Air pollution is a major global environmental
  risk to human health (WHO, 2018)
• We are simultaneously exposed to a complex
  mixture of air pollutants
• Moving toward a multi-pollutant approach to
  air quality

                     R. Argiento   Milano, 24 Settembre
Applications: Atmospheric Environment

                                           • Air pollution is a major global environmental
                                                         risk to human health (WHO, 2018)
                                           • We are simultaneously exposed to a complex
                                                         mixture of air pollutants
                                           • Moving toward a multi-pollutant approach to
                                                         air quality

                                                    4
• A better understanding of the
  interactions between air pollutants is    µg m3
                                                    2

  critical
• Learning dependencies among
                                                    0

  multiple time series
                                                    −2

                                                                                                   C6H6      NO2          SO2          O3      PM2.5      BC

                                                         2014/01/01   2014/07/02   2014/12/09   2015/05/27   2016/01/14         2016/09/20   2017/04/30    2017/10/25   2018/04/05   2018/10/11

                                                                                                                   R. Argiento                            Milano, 24 Settembre
Mixture models in Bayesian nonparametrics
Hierarchical modelling
                                                  ind.
                         Y1 , ..., Yn |θ1 , ..., θn ∼ f (yi |θi )
                                        i.i.d.
                         θ1 , ..., θn |P ∼ P
                                 d P
                         P(·) = ∞       h=1 wh δτh (·)

                                                                    R. Argiento   Milano, 24 Settembre
Mixture models in Bayesian nonparametrics
Hierarchical modelling
                                                     ind.
                         Y1 , ..., Yn |θ1 , ..., θn ∼ f (yi |θi )
                                        i.i.d.
                         θ1 , ..., θn |P ∼ P
                                 d P
                         P(·) = ∞       h=1 wh δτh (·) ∼ Norm − CRM

                                             Histogram of y
              0.4
              0.3
          Density
           0.20.1
              0.0

                    −4       −2                  0                  2                 4
                                                       y

                                                                        R. Argiento       Milano, 24 Settembre
Mixture models in Bayesian nonparametrics
Hierarchical modelling
                                                     ind.
                         Y1 , ..., Yn |θ1 , ..., θn ∼ f (yi |θi )
                                        i.i.d.
                         θ1 , ..., θn |P ∼ P
                                 d P
                         P(·) = ∞       h=1 wh δτh (·) ∼ Norm − CRM

                                             Histogram of y
              0.4
              0.3
          Density
           0.20.1
              0.0

                    −4       −2                  0                  2                 4
                                                       y

                                                                        R. Argiento       Milano, 24 Settembre
Mixture models in Bayesian nonparametrics
Hierarchical modelling
                                                     ind.
                         Y1 , ..., Yn |θ1 , ..., θn ∼ f (yi |θi )
                                        i.i.d.
                         θ1 , ..., θn |P ∼ P
                                 d P
                         P(·) = ∞       h=1 wh δτh (·) ∼ Norm − CRM

                                             Histogram of y
              0.4
              0.3
          Density
           0.20.1
              0.0

                    −4       −2                  0                  2                 4
                                                       y

                                                                        R. Argiento       Milano, 24 Settembre
Mixture models in Bayesian nonparametrics
Hierarchical modelling
                                                     ind.
                         Y1 , ..., Yn |θ1 , ..., θn ∼ f (yi |θi )
                                        i.i.d.
                         θ1 , ..., θn |P ∼ P
                                 d P
                         P(·) = ∞       h=1 wh δτh (·) ∼ Norm − CRM

                                             Histogram of y
              0.4
              0.3
          Density
           0.20.1
              0.0

                    −4       −2                  0                  2                 4
                                                       y

                                                                        R. Argiento       Milano, 24 Settembre
Mixture models in Bayesian nonparametrics
Hierarchical modelling
                                                     ind.
                         Y1 , ..., Yn |θ1 , ..., θn ∼ f (yi |θi )
                                        i.i.d.
                         θ1 , ..., θn |P ∼ P
                                 d P
                         P(·) = ∞       h=1 wh δτh (·) ∼ Norm − CRM

                                             Histogram of y
              0.4
              0.3
          Density
           0.20.1
              0.0

                    −4       −2                  0                  2                 4
                                                       y

                                                                        R. Argiento       Milano, 24 Settembre
Mixture models in Bayesian nonparametrics
Hierarchical modelling
                                                                           ind.
                                            Y1 , ..., Yn |θ1 , ..., θn ∼ f (yi |θi )
                                                                  i.i.d.
                                            θ1 , ..., θn |P ∼ P
                                                    d P
                                            P(·) = ∞       h=1 wh δτh (·) ∼ Norm − CRM

     Ongoing works

  (a) Dependent processes Px , to include covariate information
  (b) A general class of dependent models that encompasses many specific structures
  (c) Scalable algorithms for fast inference

   Selected Publications
   • Argiento, R., Cremaschi, A. and Vannucci, M. (2019). “Hierarchical Normalized Completely Random Measures to Cluster Grouped Data”, Journal
       of the American Statistical Association. Just accepted.
   •   Cremaschi, A., Argiento, R., Shoemaker, K., Peterson, C.B. and Vannucci M. (2019). “Hierarchical Normalized Completely Random Measures for
       Robust Graphical Modeling”. Bayesian Analysis. Just accepted.
   •   Argiento R., Ruggiero, M. (2018). “Computational challenges and temporal dependence in Bayesian nonparametric models”, Statistical Methods
       and Applications, Volume 27,

                                                                                                         R. Argiento        Milano, 24 Settembre
Graphical modelling

Graph theory
                                                             1                                       1
            G = (V, E)
                                      6                                6
  • finite set of vertices
    V = {1, . . . , q}                                   3                                       3
                                              5                                     5
  • subset of edges
                                                                  2                                            2
    E ⊆V ×V
                                                   4                                     4

Nodes ⇔ Random variables                  Figure: Directed (left) and undirected (right) graphs.
Edges ⇔ Probabilistic relationships

                                                                      R. Argiento       Milano, 24 Settembre
Graphical modelling

Graph theory
                                                                1                                       1
            G = (V, E)
                                         6                                6
  • finite set of vertices
    V = {1, . . . , q}                                      3                                       3
                                                 5                                     5
  • subset of edges
                                                                     2                                            2
    E ⊆V ×V
                                                      4                                     4

Nodes ⇔ Random variables                     Figure: Directed (left) and undirected (right) graphs.
Edges ⇔ Probabilistic relationships

Graphical model
3 Family of probability distributions for the q random variables which factorizes according to
a given graph. 3 Conditional independencies are read from the graph.

                                                                         R. Argiento       Milano, 24 Settembre
Graphical modelling

Graph theory
                                                                                                         1                                                1
                  G = (V, E)
                                                                        6                                                6
  • finiteOngoing
           set of vertices
                   works
      V = {1, . . . , q}                                                                            3                                                 3
                                                 5                                                                                    5
      (a) Objective Bayes Model Selection from Observational
  • subset                                                   Data
           of edges                                                                                                 2                                              2
    E (b)
      ⊆ VMultiple
           × V Graphical Models
         (c) Estimate Causal effects using Directed Graphical Models
                                                                                            4                                                4
         (c) Dependent graphs, spatio-temporal dependence to capture graphs relationships.
Nodes ⇔ Random variables                                                     Figure: Directed (left) and undirected (right) graphs.
Edges ⇔ Probabilistic relationships
          Pubblicazioni recenti
          • Castelletti, F. Consonni, G., Della Vedova, M. L. & Peluso, S. (2018). “Learning Markov equivalence classes of Directed Acyclic Graphs: an
GraphicalObjective
           model   Bayes Approach.” Bayesian Analysis 13, 1231–1256.
           •   Castelletti, F. & Consonni, G. (2019). “Objective Bayes model selection of Gaussian interventional essential graphs for the identification of
3 Family signaling   pathways.” Annals of Applied Statistics, in-press
           of probability           distributions for the q random variables which factorizes according to
a given •graph.
           Paci, L. & Consonni, G. (2019). “ Structural Learning of Contemporaneous Dependencies in Graphical VAR models”, Invited revision
                    3 Conditional independencies are read from the graph.

                                                                                                                        R. Argiento         Milano, 24 Settembre
Scientific Events

• Applied Bayesian Statistical School.
• Scientific board of BaySM-Bayesian Young
  Stitistician Meeting. Scientific chair della
  prossima edizione.
• International Society for Bayesian Analysis -
  ISBA
• Società Italiana di Statistica-SIS

                       R. Argiento     Milano, 24 Settembre
Scientific Events

    • Applied Bayesian Statistical School.
    • Scientific board of BaySM-Bayesian Young
      Stitistician Meeting. Scientific chair della
      prossima edizione.
    • International Society for Bayesian Analysis -
      ISBA
    • Società Italiana di Statistica-SIS

Lucia Paci, Alessia Pini, Raffaele Argiento,

Federico Castelletti, Stefano Peluso, Bruno

Buonaguidi, Guido Consonni

Grazie!

                           R. Argiento     Milano, 24 Settembre
You can also read