Déposez votre fichier ici pour le déplacer vers cet enregistrement.
y
Deep generative models parametrize very flexible families of distributions able to fit complicated datasets of images or text. These models provide independent samples from complex high-distributions at negligible costs. On the other hand, sampling exactly a target distribution, such as Boltzmann distributions and Bayesian posteriors is typically challenging: either because of dimensionality, multi-modality, ill-conditioning or a combination of the previous. In this talk, I will review recent works trying to enhance traditional inference and sampling algorithms with learning. I will present in particular flowMC, an adaptive MCMC with Normalizing Flows along with first applications and remaining challenges.
[-]
Deep generative models parametrize very flexible families of distributions able to fit complicated datasets of images or text. These models provide independent samples from complex high-distributions at negligible costs. On the other hand, sampling exactly a target distribution, such as Boltzmann distributions and Bayesian posteriors is typically challenging: either because of dimensionality, multi-modality, ill-conditioning or a combination of ...
[+]
68T99 ; 82B80 ; 62F15
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
y
We discuss Arithmetic Statistics as a 'new' branch of number theory by briefly sketching its development in the last 50 years. The non-triviality of the meaning of `random behaviour' and the problematic absence of good probability measures on countably infinite sets are illustrated by the example of the 1983 Cohen-Lenstra heuristics for imaginary quadratic class groups. We then focus on the Negative Pell equation, of which the random behaviour in the case of fundamental discriminants (Stevenhagen's conjecture) has now been established after 30 years.
We explain the open conjecture for the general case, which is based on equidistribution results for units over residue classes that remain to be proved.
[-]
We discuss Arithmetic Statistics as a 'new' branch of number theory by briefly sketching its development in the last 50 years. The non-triviality of the meaning of `random behaviour' and the problematic absence of good probability measures on countably infinite sets are illustrated by the example of the 1983 Cohen-Lenstra heuristics for imaginary quadratic class groups. We then focus on the Negative Pell equation, of which the random behaviour ...
[+]
11R11 ; 11R45 ; 11K99
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
y
(joint with Rideau-Kikuchi)
One of the most striking results of the model theory of henselian valued fields is the Ax-Kochen/Ershov principle, which roughly states that the first order theory of a henselian valued field that is unramified is completely determined by the first order theory of its residue field and the first order theory of its value group. Our leading question is: Can one obtain an Imaginary Ax-Kochen/Ershov principle? In previous work, I showed that the complexity of the value group requires adding the stabilizer sorts. In previous work, Hils and Rideau-Kikuchi showed that the complexity of the residue field reflects by adding the interpretable sets of the linear sorts. In this talk we present recent results on weak elimination of imaginaries that combine both strategies for a large class of henselian valued fields of equicharacteristic zero. Examples include, among others, henselian valued fields with bounded galois group and henselian valued fields whose value group has bounded regular rank (with an angular component map).
[-]
(joint with Rideau-Kikuchi)
One of the most striking results of the model theory of henselian valued fields is the Ax-Kochen/Ershov principle, which roughly states that the first order theory of a henselian valued field that is unramified is completely determined by the first order theory of its residue field and the first order theory of its value group. Our leading question is: Can one obtain an Imaginary Ax-Kochen/Ershov principle? In previous work, ...
[+]
03C60
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
y
The Ceresa class is the image under a cycle class map of a canonical algebraic cycle associated to a curve in its Jacobian. This class vanishes for all hyperelliptic curves, and is known to be non-vanishing for the generic curve of genus at least 3. It is necessary for the Ceresa class to have infinite order for the Galois action on the fundamental group of a curve to have big image. We will present an algorithm for certifying that a curve over a number field has infinite order Ceresa class.
N.B. This is preliminary joint work with Jordan Ellenberg, Adam Logan and Akshay Venkatesh.
[-]
The Ceresa class is the image under a cycle class map of a canonical algebraic cycle associated to a curve in its Jacobian. This class vanishes for all hyperelliptic curves, and is known to be non-vanishing for the generic curve of genus at least 3. It is necessary for the Ceresa class to have infinite order for the Galois action on the fundamental group of a curve to have big image. We will present an algorithm for certifying that a curve over ...
[+]
14C25 ; 14H25 ; 14Q05 ; 11G30 ; 11G40
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
Déposez votre fichier ici pour le déplacer vers cet enregistrement.
y
Privacy concerns are becoming a major obstacle to using data in the way that we want. It's often unclear how current regulations should translate into technology, and the changing legal landscape surrounding privacy can cause valuable data to go unused. How can data scientists make use of potentially sensitive data, while providing rigorous privacy guarantees to the individuals who provided data? A growing literature on differential privacy has emerged in the last decade to address some of these concerns. Differential privacy is a parameterized notion of database privacy that gives a mathematically rigorous worst-case bound on the maximum amount of information that can be learned about any one individual's data from the output of a computation. Differential privacy ensures that if a single entry in the database were to be changed, then the algorithm would still have approximately the same distribution over outputs. In this talk, we will see the definition and properties of differential privacy; survey a theoretical toolbox of differentially private algorithms that come with a strong accuracy guarantee; and discuss recent applications of differential privacy in major technology companies and government organizations.
[-]
Privacy concerns are becoming a major obstacle to using data in the way that we want. It's often unclear how current regulations should translate into technology, and the changing legal landscape surrounding privacy can cause valuable data to go unused. How can data scientists make use of potentially sensitive data, while providing rigorous privacy guarantees to the individuals who provided data? A growing literature on differential privacy has ...
[+]
68W40 ; 68-02 ; 62-02 ; 90-02