University of Southern California

Diffusion models, which transform noise into new data instances by reversing a Markov diffusion process, have become a cornerstone in modern generative models. A key component of these models is to learn the score function through score matching. While the practical power of diffusion models has now been widely recognized, the theoretical developments remain far from mature. Notably, it remains unclear whether gradient-based algorithms can learn the score function with a provable accuracy. In this talk, we develop a suite of non-asymptotic theory towards understanding the data generation process of diffusion models and the accuracy of score estimation. Our analysis covers both the optimization and the generalization aspects of the learning procedure, which also builds a novel connection to supervised learning and neural tangent kernels.

This is based on joint work with Yinbin Han and Meisam Razaviyayn (USC).

Beginning with the foundational work of Daniel Quillen, an understanding of aspects of the cohomology of finite groups evolved into a study of representations of finite groups using geometric methods of support theory. Over decades, this approach expanded to the study of representations of a vast array of finite dimensional Hopf algebras. I will discuss how related geometric and categorical techniques can be applied to linear algebra groups.

The subject of “geometric” fluid dynamics flourished following the seminal work of VI.
Arnold in the 1960s. A famous paper was published in 1970 by David Ebin and Jerrold
Marsden, who used the manifold structure of certain groups of diffeomorphisms to obtain
sharp existence and uniqueness results for the classical equations of fluid dynamics. Of
particular importance are the fixed points of the underlying dynamical system and the
“accessibility” of these Euler equilibria. In 1985 Keith Moffatt introduced a mechanism
for reaching these equilibria not through the Euler vortex dynamics itself but via a
topology-preserving diffusion process called “Magnetic Relaxation”. In this talk, we will
discuss some recent results for Moffatt’s MR equations which are mathematically
challenging not only because they are active vector equations but also because they have
a cubic nonlinearity.

This is joint work with Rajendra Beckie, Adam Larios, and Vlad Vicol.

In the event of food-borne disease outbreaks, conventional epidemiological approaches to identify the causative food product are time-intensive and often inconclusive. Data-driven tools could help to reduce the number of products under suspicion by efficiently generating food-source hypotheses. We frame the problem of generating hypotheses about the food-source as one of identifying the source network from a set of food supply networks (e.g. vegetables, eggs) that most likely gave rise to the illness outbreak distribution over consumers at the terminal stage of the supply network. We introduce an information-theoretic measure that quantifies the degree to which an outbreak distribution can be explained by a supply network’s structure and allows comparison across networks. The method leverages a previously-developed food-borne contamination diffusion model and probability distribution for the source location in the supply chain, quantifying the amount of information in the probability distribution produced by a particular network-outbreak combination. We illustrate the method using supply network models from Germany and demonstrate its application potential for outbreak investigations through simulated outbreak scenarios and a retrospective analysis of a real-world outbreak.

Abstract: In this talk we try to establish an analytic framework for studying Set-Valued Backward Stochastic Differential Equations (SVBSDE for short), motivated largely by the current studies of dynamic set-valued risk measures for multi-asset or network-based financial models. Our framework will be based on the notion of Hukuhara difference between sets, in order to compensate the lack of “inverse” operation of the traditional Minkowski addition, whence the vector space structure, in traditional set-valued analysis. We shall examine and establish a useful foundation of set-valued stochastic analysis under this algebraic framework, including some fundamental issues regarding Aumann-Itˆo integrals, especially when it is connected to the martingale representation theorem. We shall identify some fundamental challenges and propose some extensions of the existing theory that are necessary to study the SVBSDEs. This talk is based on the joint works with C¸ a˘gın Ararat and Wenqian Wu.

The finance literature documents a relation between labor income and

the cross-section of stock returns. One possible explanation for this

is the hedging decisions of investors with relative wealth concerns.

This implies a negative risk premium associated with stock returns

correlated with local undiversifiable wealth, since investors are

willing to pay more for stocks that help their hedging goals. We find

evidence that is consistent with these regularities. In addition, we

show that the effect varies across geographic areas depending on the

size and variability of undiversifiable wealth, proxied by labor income.

Factorization homology is an invariant of an n-manifold M together with an n-disk algebra A. Should M be

a circle, this recovers the Hochschild complex of A; should A be a commutative algebra, this recovers the

homology of M with coefficients in A. In general, factorization homology retains more information about

a manifold than its underlying homotopy type.

In this talk we will lift Poincare' duality to factorization homology as it intertwines with Koszul

duality for n-disk algebras -- all terms will be explained. We will point out a number of consequences

of this duality, which concern manifold invariants as well as algebra invariants.

This is a report on joint work with John Francis.

We consider an active scalar equation with singular drift velocity that is motivated by a model for the geodynamo. We show that the non-diffusive equation is ill-posed in the sense of Hadamard in Sobolev spaces. In contrast, the critically diffusive equation is globally well-posed. This work is joint with Vlad Vicol.