Data-Adaptive Statistical Inference: foreword to the DGIJB special issue
Résumé
The concomitant emergence of big data, explosion of ubiquitous computational
resources and democratization of the access to more powerful computing make it
necessary and possible to rethink pragmatically the practice of statistics.
While numerous machine learning methods provide much ever easier access to
data-mining tools and sophisticated prediction, there is a growing realization
that ad hoc and non-prespecified approaches to high-dimensional problems lend
themselves to a proliferation of ``findings'' of dubious reproducibility.
This period of fast-paced evolution is thus a blessing for statistics. It is a
golden opportunity to build upon more than a century of methodological
research in statistics and five decades of methodological research in machine
learning to bend the course of statistics in a new direction, away from the
misuse of parametric models and reporting of non-robust inference, to tackle
rigorously the challenges that we, as a community, are confronted with.
We asked researchers currently engaged in cutting edge research on
data-adaptive inferential methods to share their views with us. The result is
a compelling collection of advances in statistical theory and practice.
Domaines
Statistiques [math.ST]Origine | Fichiers produits par l'(les) auteur(s) |
---|