Optimization and Monte Carlo Sampling Intertwined – OptiMoCSI

Elaborating on earlier works dedicated to developing advanced optimization-based signal processing tools for the monitoring of the time-space evolution of the Covid19 pandemic intensity (quantified by the reproduction number) in a crisis / emergency context and from limited quality data, the main objectives of OptiMoCSI are to develop methodological contributions, combining (nonsmooth) optimization and Monte Carlo sampling, to address challenges posed in high societal stakes problems.
First, OptiMoCSI will devise (self-exciting type) statistical frameworks for the joint modeling of application-driven real-world mechanisms and data or observations, often consisting of non-stationary and non-negative counts highly corrupted by outliers or missing values.
Second, OptiMoCSI will devise tools permitting theoretically sound and practically efficient statistical (Bayesian Monte Carlo type) estimation of the parameters of interest and, importantly, yielding uncertainty assessment.
Third, OptiMoCSI will construct (deterministic and stochastic inference based) tools permitting to assess the robustness of achieved estimations with respect to arbitrary choices in either real-world mechanism or observation quality modeling.
Indeed, in addressing, with science-based tools, high societal stake problems entailing decision making and/or citizen information, uncertainty and robustness assessments in estimated outputs constitute both mandatory, critical and moral duties as well as significant technical scientific challenges.
Fourth, in an attempt to contribute to open/reproducible science and to facilitate the relationship between science and society, OptiMoCSI will invent cartography-based visualization tools that enable better grasp and thus better use of the notion of uncertainty both for interdisciplinary research deployments and for communication with general non-scientific audiences.
Beyond methodological ambitions, OptiMoCSI intends to produce a set of documented toolboxes oriented towards achieving three main objectives.
In collaboration with epidemiologists, the real and systematic implementation on Covid19 pandemic data should allow to evaluate if and how the computational tools developed go significantly beyond the state of the art in quantitative epidemiology, especially in emergency contexts (limited quality data), or immediately at the onset of the pandemic, or for the adaptation to other types of epidemics (influenza, ebola,...)
Beyond epidemiology, the methodological tools and toolboxes are intended to be of generic use for a wide variety of applications in signal and image processing, where an underlying self-exciting process is observed through an imperfect counting device. The expected results could also extend to a wider variety of cases where a variable intensity function needs to be estimated in a low count regime, e.g. photonics or Lidar imaging techniques.
In addition, the toolboxes will be designed to be used by non-experts in statistical signal processing, in other interdisciplinary research projects in social sciences and humanities (e.g., focused on virality in social networks or detection of re-sharing cascades), or oriented towards high-stakes societal challenges, policy makers, or citizen information.