Date: 10/1/2023

Speaker: Sebastian Ferrando (Toronto Metropolitan University)

Title: Non-probabilistic supermartingales

Abstract: We provide a brief motivational overview of recent developments of extensions of stochastics tools to deal with uncertainty. These are: Peng's nonlinear expectations and Ito's calculus without probabilities. We then describe a non-probabilistic version of a supermartingale theory closely motivated by financial considerations of no-arbitrage. The basic object replacing the classical filtered probability space is a structured trajectory set which allows the definition of conditional outer integrals as well as null sets. The conditional outer integrals are non linear functionals that allow to circumvent the linearity of the classical conditional expectations in proofs and definitions. Integrability notions emerge in our setting through non-classical conditional integral operators that lead to the special case of non-probabilistic martingales. One can define non-probabilistic supermartingales and prove analogous of classical results like: Doob's optional sampling theorem, Dubin's upcrossing inequalities and Doob's a.e. convergence for non-negative supermartingales. All constructions and results have a hedging and superhedging interpretation and there is a direct way in which the new results generalize the classical case. Null sets appearing in the results have a financial interpretation and are handled in a more concrete way than in the classical theory.

Date: 17/1/2023

Speaker: Gourab Ray (UVic)

Title: Characterizing nonamenability through stochastic domination and finitary factors

Abstract: Take an Ising model with very low temperature. What is the largest p such that the Ising model dominates Bernoulli percolation with parameter p ? We will show that the answer to this question depends drastically on the geometry of the graph. We also obtain similar results for for two Ising models at very low, but close temperatures. A process is a finitary factor of iid if it can be written as a measurable and equivariant function of an iid process. As an application of the domination results, we show that the very low temperature Ising model on a nonamenable graph is a finitary factor of iid. This is in stark contrast with the amenable setting, where it is known through a celebrated result of Van Den Berg and Steif that the low temperature Ising model is not a finitary factor of iid. Joint work with Yinon Spinka.

Date: 24/1/2023

Speaker: Grigory Terlov (Illinois)

Title: Stein’s method for conditional central limit theorem

Abstract: It is common in probability theory and statistics to study distributional convergences of sums of random variables conditioned on another such sum. In this talk I will present a novel approach using Stein’s method for exchangeable pairs that allows to derive a conditional central limit theorem of the form $(X_n|Y_n = k)$ with explicit rate of convergence as well as its extensions to a multidimensional setting. We will apply these results to particular models including pattern counts in a random binary sequence and subgraph counts in Erdös-Rényi random graph. This talk is based on joint work with Partha S. Dey.

Date: 31/1/2023

Speaker: Will Perkins (Georgia Tech)

Title: The (symmetric) Ising perceptron: progress and problems

Abstract: The Perceptron model was proposed as early as the 1950's as a toy model of a one-layer neural network. The basic model consists of a set of solutions (either the Hamming cube or the sphere of dimension n) and a set of constraints given by independent n-dimensional Gaussian vectors. The constraints are that the inner product of a solution vector with each constraint vector scaled by sqrt{n} must lie in some interval on the real line. Probabilistic questions about the model include the satisfiability threshold (or the "storage capacity") and questions about the typical structure of the solution space. Algorithmic questions include the tractability of finding a solution (the learning problem in the neural network interpretation). I will describe the model, the main problems, and recent progress.

DATE: 7/2/2023

Speaker: Peleg Michaeli (Carnegie Mellon)

Title: Fast construction on a restricted budget

Abstract: We introduce a model of a controlled random process. In this model, the vertices of a hypergraph are ordered randomly and then revealed, one by one, to an algorithm. The algorithm must decide, immediately and irrevocably, whether to keep each observed vertex. Given the total number of observed vertices ("time"), the algorithm's goal is to succeed - asymptotically almost surely - in completing a hyperedge by keeping ("purchasing") the smallest possible number of vertices. We analyse this model in the context of random graph processes, where the corresponding hypergraph defines a natural graph property, such as minimum degree, connectivity, Hamiltonicity and the containment of fixed-size subgraphs. Joint work with Alan Frieze and Michael Krivelevich.

Date: 14/2/2023

Speaker: Yin-Ting Liao (Irvine)

Title: Large deviations for projections of high-dimensional measures

Abstract: Random projections of high-dimensional probability measures have gained much attention in asymptotic convex geometry and high-dimensional statistics. While fluctuations at the level of the central limit theorem have been classically studied, only more recently has an inquiry into large deviation principles for such projections been initiated. In this talk, I will review existing work and describe our results on large deviations. I will also talk about sharp large deviation estimates to obtain the prefactor apart from the exponential decay in the spirit of Bahadur and Ranga-Rao. Applications to asymptotic convex geometry and a range of examples including \ell^p balls and Orlicz balls would be given. This talk is based on several joint works with S. S. Kim and K. Ramanan.

Date: 28/2/2023

Speaker: Ahmed Bou-Rabee (Cornell)

Title: Random growth on a random surface

Abstract: I will describe the large-scale behavior of a random growth model (Internal DLA) on random planar maps which approximate a random fractal surface embedded in the plane (Liouville quantum gravity, LQG). No prior knowledge of these objects will be assumed. Joint work with Ewain Gwynne.

Date: 7/3/2023

Speaker: Máté Wierdl (Memphis)

Title: Generation of measures by statistics of rotations along sets of integers

Abstract: Let S be a strictly increasing positive integer sequence s_1

Speaker: Pablo Shmerkin (UBC)

Title: Incidences and line counting: from the discrete to the fractal setting

Abstract: How many lines are spanned by a set of planar points?. If the points are collinear, then the answer is clearly "one". If they are not collinear, however, several different answers exist when sets are finite and "how many" is measured by cardinality. I will discuss a bit of the history of this problem and present a recent extension to the discretized and continuum settings, obtained in collaboration with T. Orponen and H. Wang. No specialized background will be assumed.

Date: 21/3/2023

Speaker: Neha Bansal (UBC Okanagan)

Title: Reproductive value for time homogenous branching population models

Abstract: Reproductive value is the relative expected number of off-springs produced by an individual in its remaining lifetime. It is also an invariant function for population processes with birth and death rates independent of the time except in cases when they are periodic. In this study, we prove that the limiting ratio of the survival probability of two branching processes starting from distinct state-time positions is equal to the relative reproductive value. Moreover, we developed a method to obtain the reproductive value for continuous time general branching population models with state-dependent rates and a renewal state. We are using size-biased birth rates for constructing the spinal representation of the branching process, which establishes the relation between survival probability and the Martingale of the process. Further, we provide sufficient conditions for a successful coupling to compare the survival probabilities of two branching population models.

Date: 28/3/2023

Speaker: Emilio Corso (UBC)

Title: Randomness of flows in negative curvature

Abstract: A momentous legacy of twentieth-century mathematics is the realisation that deterministically evolving systems frequently exhibit, when observed for sufficiently extended periods of time, a statistical behaviour akin to the limiting behaviour of independent random variables. We shall explore a geometric incarnation of this surprising phenomenon, overviewing various kinds of statistical limit theorems for the free motion of a particle on a negatively curved surface. In order to emphasise the richness of possible asymptotic behaviours, as well as the variety of sources of randomness, we will further compare the free-motion dynamics with a closely related evolution on the same phase space, known as the horocycle flow.

Date: 4/4/2023

Speaker: Natalie Behague (UVic)

Title: Probabilistic Zero Forcing on Hypercubes and Grids

Abstract: Probabilistic zero-forcing can be thought of as a model for rumour spreading, where a person is more likely to spread a rumour if several of their friends already believe it . We start with a graph that has one infected vertex. At each time step an infected vertex infects an uninfected neighbour with probability proportional to how many of its neighbours are already infected. I will focus in this talk on probabilistic zero-forcing on hypercubes and grids, and demonstrate tight bounds on how long it takes for every vertex to be infected (asymptotically almost surely). This is joint work with Trent Marbach and Pawel Pralat.

For (some) previous semesters, see

Spring 2006 |

Fall 2006 |

Spring 2007 |

Spring 2008 |

Spring 2009 |

Spring 2010 |

Spring 2011 |

Spring 2012 |

Spring 2013 |

Spring 2014 |

Spring 2015 |

Autumn 2020 |

Spring 2021 |

Spring 2022 |