Product-Form Probability Distributions
The workshop Product-Form Probability Distributions, which will take place from 20-23 May 2025, will focus on “product-form” probability distributions, where the mass or density function is expressed as a product of factors, each depending on the sample and model parameters.
These distributions appear naturally in many fields such as queuing theory, statistical physics, or machine learning. For example, in the analysis of Markov chains and Markov decision processes, the stationary distribution sometimes has a product-form, as a consequence of the balance equations. Other classical examples include Jackson networks in queueing theory, the Ising model and zero-range process in statistical physics, and exponential families in data science, which arise naturally as maximum-entropy distributions.
Product-form distributions also enable efficient inference algorithms, for example, to calculate the normalizing constant or sufficient statistics. In queueing theory, examples include the Buzen algorithm and mean-value analysis for Jackson networks. In machine learning, particularly in probabilistic graphical models, sum-product (or message-passing) algorithms leverage the network structure for efficient inference. In queueing theory and statistical physics, product-form distributions are also exploited to help understand phase transitions in various scaling regimes.
The goal of this workshop is threefold:
- review the state of the art in product-form distributions,
- promote an interdisciplinary approach to product-form distributions,
- identify and highlight research questions and opportunities in this domain.
Organizers
- Céline Comte (CNRS & LAAS, France)
- Agnieszka Janicka (Eindhoven University of Technology, The Netherlands)
- Richard Boucherie (University of Twente, The Netherlands)
More information and registration
More information and the registration form can be found on the website of EURANDOM
