Working Groups

Working Groups are an integral part of SAMSI’s research programs, where participants explore research themes determined in the Opening Workshop.

Program on Numerical Analysis in Data Science (Fall 2020)

  • Working Group I:  Large-scale Inverse Problems and Uncertainty Quantification

The focus of this working group is on advancing computational tools for large-scale inverse problems and uncertainty quantification.  We tackle a range of open problems from the development of new regularization approaches for computing solutions to inverse problems to the advancement of technologies for large-scale UQ.  We aim to bring together researchers in numerical analysis, probability and statistics, and domain experts with connections to applications of inverse problems.

  • Working Group II:  Global Sensitivity Analysis

How complex should a computational model be to be useful? Both the mathematical community and domain scientists have long been worried about the lower bound, i.e., minimum complexity. This working group will consider the problem of finding the lowest level complexity sufficient for specific tasks. The working group will gather scientists who are making fundamental contributions to computational modeling under uncertainty through the development of mathematical theory and algorithms for sensitivity analysis of stochastic models as well as goal-oriented dimension reduction for fast UQ and parameter estimation.

  • Working Group III:  Randomized Algorithms for Matrices and Data

Randomized Numerical Linear Algebra (RandNLA)  is an interdisciplinary area which uses randomization as a computational tool for tackling large linear algebra problems. While it has origins in theoretical computer science, RandNLA has connections to many areas of applied and computational mathematics. This working group will bring together researchers in numerical analysis, theoretical computer science, scientific computing, machine learning, statistics, and domain experts with connections to RandNLA. The goal is to not only address foundational questions in RandNLA but to explore and investigate the applications of RandNLA algorithms to new areas in scientific computing and data science including inverse problems, uncertainty quantification, model reduction, and tensor decompositions.

  • Working Group IV:  Computational Algorithms for Reinforcement Learning

Due to many successful applications in robotics, games, precision health, e-commerce and ride-sharing industries, Reinforcement Learning (RL) has gained great popularity among various scientific fields. The goal of this working group is to join forces of researchers from different academic fields and industry, to explore cutting edge algorithms, applications and theory, including and not limited to deep RL, model-based RL, multi-agent RL, inverse RL and policy evaluation methods.

Online seminars will be presented detailing the latest advances in reinforcement learning applications and theory. The goal is to bring virtual seminars featuring the latest work in applying reinforcement learning methods in many exciting areas (e.g., health sciences, or two-sided markets).  Click here for details:

  • Working Group V:  Dimension Reduction in Time Series

Studies in sufficient dimension reduction for multivariate time series, focusing up Bayesian methods to estimate the central mean subspace, principal component analysis, factor models, and envelope methods.

Program on Combinatorial Probability (Spring 2021)

There is an interest in extending random processes on graphs to simplicial complexes. This working group will explore problems in this area

The working group will explore the use of ideas from group theory and representation theory  in studying the properties of random combinatorial objects. This can include distributions over random groups and using representation theory to study random combinatorial objects such as rankings or Young tableaus.

This working group will explore the relationship between phase transitions in statistical physics models (e.g. Ising, Potts, hard-core) and efficient algorithms for approximate counting and sampling. While this is a huge field with many different techniques and connections to other areas of mathematics and computer science, we will focus on three specific questions that have emerged in the last few years: 1) when can we develop algorithms that work in the low-temperature, strong interaction regime of a statistical physics model? 2) what are the connections between the three main approaches to approximate counting: Markov chain Monte Carlo, the method of correlation decay, and the polynomial interpolation method?  3) how can we use the worst-case, computational perspective to better understand phase transitions in physics models? The working group will read lecture notes and learn the area as we proceed, so no background in statistical physics is necessary.

Program on Data Science in the Social and Behavioral Sciences (Spring 2021)

  • Working Group I: TBD
  • Working Group II:  TBD
  • Working Group III: TBD

Additional information for Working Group Leaders and Web Masters is provided HERE

Past Working Groups