[ABE-L] COLMEA - dia 15 de junho (quarta-feira) Ã s 14h - PUC-Rio

[Maria Eulalia Vares]

Prezados colegas,

Nosso próximo encontro do COLMEA terá lugar na PUC-Rio, no próximo dia 15.
Será o primeiro encontro presencial depois de fevereiro de 2020.  Na
ocasião teremos a seguinte programação:

*14:00 h* - *Paulo Orenstein (IMPA)*

*Split conformal prediction and its extensions to non-exchangeable data*

*15:20 h - Ivailo Hartarsky (Université Paris-Dauphine)*

*Universality in bootstrap percolation and kinetically constrained models*

Todos são muito bem-vindos.

*Local do evento: *Sala de reuniões do Decanato do CTC

12 º andar do prédio Cardeal Leme

PUC-Rio, Gávea

Mais informações sobre o COLMEA podem ser encontradas através da homepage
http://www.im.ufrj.br/~coloquiomea/

Os resumos das palestras estão no final da mensagem.  Agradecemos se puder
divulgar. Em anexo o cartaz de divulgação.

Atenciosamente,

O comitê organizador:

Americo Cunha (UERJ)

Augusto Q. Teixeira (IMPA)

Evaldo M. F. Curado (CBPF)

João Batista M. Pereira (UFRJ)

Leandro P. R. Pimentel (UFRJ)

Maria Eulalia Vares (UFRJ)

Nuno Crokidakis (UFF)

Simon Griffiths (PUC-Rio)



--------------

*Resumos das palestras*

*Split conformal prediction and its extensions to non-exchangeable data*

*Paulo Orenstein (IMPA)*

Machine learning algorithms offer state-of-the-art predictive performance
in a variety of domains, but often lack an associated measure of
uncertainty quantification. Split conformal prediction is a leading tool to
obtain predictive intervals with virtually no assumptions beyond data
exchangeability. This crucial assumption, however, hinders its
applicability to many important data, such as time series and spatially
dependent processes. In this talk, we will introduce split CP and show how
it can be extended to non-exchangeable settings through a small coverage
penalty. The proposed framework, based on concentration of measure
inequalities, works more generally than traditional split CP, and
experiments corroborate our coverage guarantees even under highly dependent
data.

This is joint work with Roberto Imbuzeiro Oliveira, Thiago Ramos and João
Vitor Romano.


*Universality in bootstrap percolation and kinetically constrained models*

*Ivailo Hartarsky (Université Paris-Dauphine)*

The paradigmatic 2-neighbour bootstrap percolation model is the following
cellular automaton. Given a set of infected sites in Z^d, we iteratively
infect each site with at least 2 infected neighbours, while infections
never heal. We are then interested in whether and when the origin becomes
infected under this dynamics starting from an i.i.d. Bernoulli initial
infection. There is a naturally associated stochastic non-monotone model:
the Fredrickson-Andersen 2-spin facilitated one, in which the state of each
site is resampled to a Bernoulli variable at rate 1, provided it has at
least 2 infected neighbours.

Of course, many related models have been considered, replacing the
2-neighbour constraint by an increasing local translation-invariant
constraint (e.g. both the North and East neighbours are infected). In this
talk we will overview recent universality results for this class of
bootstrap percolation and its non-monotone stochastic counterpart called
kinetically constrained models. The outcome is a classification of all
rules in terms of the scaling of the infection time of the origin when the
density of infections approaches a possibly degenerate critical value.
-------------- Pr�xima Parte ----------
Um anexo em HTML foi limpo...
URL: <http://lists.ime.usp.br/pipermail/abe/attachments/20220607/984ac643/attachment-0001.htm>
-------------- Pr�xima Parte ----------
Um anexo não-texto foi limpo...
Nome: 2022_06.pdf
Tipo: application/pdf
Tamanho: 910585 bytes
Descrição: não disponível
URL: <http://lists.ime.usp.br/pipermail/abe/attachments/20220607/984ac643/attachment-0001.pdf>