Date/Time : 08 May 2024 17:00
Venue : S17 #04-04 (SR3)
Speaker : Vittorio Cipriani, Vienna University of Technology

In this talk, we combine computable structure theory and inductive inference to study learning of families of structures. All the structures we consider are relational and countable and all the families of structures we consider are at most countable. The framework we use was defined in a series of papers by Bazhenov, Fokina, Kotzing, and San Mauro. In a nutshell, the framework models the scenario in which, given a family of structures K, a learner receives more and more information about the atomic diagram of a copy of some A ∈ K and, at each stage, is required to output a conjecture about the isomorphism type of such a structure. In this context, a natural criterion to consider is Ex-learning in which we require the learner to stabilize to the correct conjecture after finitely many steps.

Together with Bazhenov and San Mauro we gave a descriptive set-theoretic characterization of Ex-learning. Namely, we showed that a family of structures is Ex-learnable if and only if the corresponding isomorphism problem continuously reduces to E_0, the equivalence relation of eventual agreement on infinite binary sequences. Replacing E0 with other equivalence relations, one obtains a hierarchy to rank such isomorphism problems. That is, a family of structures K is E-learnable, for an equivalence relation E, if there is a continuous reduction from the isomorphism problem associated with K to E.

We aim to obtain model-theoretic characterization of E-learning for different equivalence relations E. Some characterizations are already present in the literature: here we show that a family of structures K such that for any A_i, A_j ∈ K there is a Σ_n^{inf} formula satisfied by A_i but not by A_j is E-learnable if and only if E is the (iteration of the) Friedman-Stanely jump of the identity either on natural numbers or on Cantor space. We also show that other learning criteria coming from the classical setting of inductive inference of formal languages or recursive functions have a nice model-theoretic characterization.

This talk collects joint works with Bazhenov, Jain, Marcone, San Mauro and Stephan.

Add to calendar :

Date/Time : 02 May 2024 14:00
Venue : S17 #06-11 (SR6)
Speaker : Xuecheng Wang, Tsinghua University

The 3D relativistic Vlasov-Maxwell system is one of the fundamental models in the plasma physics, which describes the dynamics of electron and ions under the electromagnetic field created by particles themselves. In this talk, we discuss the existence of global solution and the propagation of regularity of the 3D relativistic Vlasov-Maxwell system for both the small initial data case and the large initial data case.

Add to calendar :

Date/Time : 02 May 2024 15:00
Venue : S17 #06-11 (SR6)
Speaker : Pengyu Le, BIMSA

In this talk, we shall introduce the Lorentz polarisation in the Minkowski spacetime, which is the analogy of the polarisation in the Euclidean space. Similar to the application of the Euclidean polarisation to the classical isoperimetric inequality, we shall apply the Lorentz polarisation to study an analogous isoperimetric inequality for the volume of domain of dependence in Minkowski spacetime, and some other geometric variational problems.

Add to calendar :

Date/Time : 24 Apr 2024 14:00
Venue : S17 #06-11 (SR6)
Speaker : Arick Shao, Queen Mary University of London

We consider heat operators on a bounded domain, with a critically singular potential diverging as the inverse square of the distance to the boundary of the domain. We establish the first boundary control results for such operators in all spatial dimensions.

First, we prove null controllability for a range of parameters representing the strength of the singular potential, and with control imposed on the entire boundary. Next, we prove a weaker approximate control result, but for the full range of parameters with a viable well-posedness theory, and with the control localised to any open subset of the boundary.

This is joint work with Alberto Enciso (ICMAT) and Bruno Vergara (Brown).

Add to calendar :

Date/Time : 23 Apr 2024 16:00
Venue : S8 #03-14
Speaker : Philipp Kunde, Jagiellonian University

One of the oldest and most influential problems in ergodic theory is the classification of invertible measure-preserving transformations (MPT’s) up to isomorphism or Kakutani equivalence. Starting from the late 1990s, so-called anti-classification results use concepts from descriptive set theory to show precisely that classification for ergodic MPT’s is impossible in terms of computable invariants, even with a very inclusive understanding of “computability”. In this talk, we present more details about our techniques to prove such anti-classification results.

In the search for classification or anti-classification results, we also place increasingly more stringent conditions on the family of transformations being considered for classification. Bernoulli shifts, which can be viewed as the most random type of transformations, have the strongest type of classification, a complete Borel invariant (namely measure-theoretic entropy). But what about classes of transformations that are more random than ergodic transformations, but less so than Bernoulli shifts, such as weakly mixing transformations, mixing transformations, and K-automorphisms?

Add to calendar :

Date/Time : 22 Apr 2024 13:30
Venue : S17 #04-06 (SR1)
Speaker : Kevin Buzzard, Imperial College London

Computer programs which understand the axioms of mathematics have existed for decades. However, it is only in the last few years that they have captured the imagination of mathematicians. In this talk, I will give an overview of where we are in 2024, what these systems can currently do, and what they cannot yet do. I will discuss how these systems have helped mathematicians, and ways in which they may help mathematicians in the future. I will be careful to distinguish facts from science fiction. The talk will be suitable for a general mathematical audience and will not assume any expertise in AI, computer science, or interactive theorem provers.

Add to calendar :

Date/Time : 19 Apr 2024 10:00
Venue : S17 #04-06 (SR1)
Speaker : Riccardo Brasca, Université Paris Cité

Workshop on Formal Proofs and Lean

Add to calendar :

Date/Time : 18 Apr 2024 16:00
Venue : S17 #05-12 (SR4)
Speaker : Teruhisa Koshikawa , Kyoto University

I will talk about some relationships among categorical Langlands, L-packets, A-packets (or rather ABV-packets), and eigensheaves, focusing on the case of GL_2.

Add to calendar :

Date/Time : 18 Apr 2024 16:30
Venue : S14 #06-19
Speaker : Ching Wei Ho, Academia Sinica

It is a classical problem to study the evolution of roots of polynomials under application of a differential operator. In this talk, I will discuss the heat evolution of random polynomials with a rotationally invariant root distribution on the complex plane. The limiting root distribution of the heat-evolved random polynomial can be completely determined in terms of its log potential. For example, when a Weyl polynomial, whose root distribution converges to the uniform distribution on the unit disk, undergoes heat flow, the limiting root distribution is uniform on some ellipse until time 1 at which it becomes exactly the semicircle law. This is joint work with Brian Hall, Jonas Jalowy, and Zakhar Kabluchko.

Add to calendar :