Graduate Students Seminar

Session Chair:    Vishal Subedi
Discussant:         Dr. Yehenew Kifle

Speaker 1:     Upama Paul Chowdhury

Title A review of Permutation accelerates Approximate Bayesian Computation (Perm-ABC)

Abstract
In the presentation, I will discuss the work titled "Permutation accelerates Approximate Bayesian Computation" by Luciano et al. As a "likelihood-free" (When likelihood functions are intractable or computationally prohibited) method, Approximate Bayesian computation (ABC) is one of the most popular choices. However, to use this method, scalability is a major challenge in hierarchical or high-dimensional models. In this paper, the authors introduce "Perm-ABC", a new framework designed for settings with global and local parameters, where observations are grouped into exchangeable compartments. Perm-ABC exploits the exchangeability of different groups through permutation-based matching, which significantly improves computational efficiency over the existing methods.

Through synthetic and real-world experiments of a hierarchical Susceptible-Infectious-Recovered model of the early COVID-19 epidemic across 94 French departments, they demonstrated the practical gains in accuracy and efficiency achieved by Perm-ABC.

Speaker 2:     Garvin Melles

Title Moment Growth Bounds on Population Processes

Abstract
Stochastic chemical reaction networks can be modeled by continuous time Markov processes with a state space of the non-negative integers. Such processes also have other applications, including in models of predator-prey systems and infectious diseases. Because it is often impractical to compute the probabilities in such systems, it is often useful to study the moments of the processes. We present two sufficient conditions for moments to satisfy an exponential growth bound.