As evidenced by historical successes, modelling was once central to the development of new polymer materials and processing routes. However, despite an increase of oders of magnitude in computational power over the past two decades, the role of modelling as a predictive tool for guiding polymer synthesis has arguably declined. The resulting chasm between modelling and experiments means that the design of new or alternatively-sourced polymeric materials remains a matter of trial and error and, in contrast to the increased adoption of machine learning (ML) in other fields of material science,the development and exploitation of digital tools in polymer science has only just started.[1]  Recognizing the urgency of the challenge, the US Materials Genome Initiative has recently expanded to include polymer science [2] but has focused on building databases of mainly experimental data and the ML tools to mine them. The National Institute for Materials Science in Japan is also supporting activities in the area of experimental data collection and the development of a national database, PolyInfo. In Europe, however, activities around data-centric approaches in polymer science, known as polymer informatics, are scattered.

Molecular simulations can help in obtaining high-quality standardised and reproducible data to build ML models for polymeric materials. [3] Computationally, polymer simulations are however, notoriously challenging due to their unique hierarchical structure that makes them the prototypical example of a multiscale modelling problem. Currently, multiscale modelling of polymers is mostly tackled using single-scale methods, which are coupled together in ad-hoc ways that are grossly inefficient, and a general conceptual framework is lacking.[4] ML methods are perfectly suited to develop a coherent modelling framework that can bring together physics-based simulations at different scales and be quickly deployed over a large number of polymeric systems.[4-7]

^{In this workshop, we will bring together leading scientists involved in modelling and data collection of polymers to discuss (and possibly agree on) how in-silico data can be efficiently harvested and effectively used for building ML models for these ubiquitous materials.  The workshop will focus on three areas:}

^{1) High-throughput simulations: What hardware and software are needed to collect data from classical simulations for polymeric systems? Can we develop community-agreed protocols to perform polymer simulations at speed?}

^{2) data representation and reproducibility: What is the most efficient machine-readable format to represent a polymeric system that can include topological as well as chemical features and be used to create input files for molecular simulations? How can we guarantee the reproducibility of the simulation results in a field characterized by many different simulation protocols and parameters? Can we devise a data representation strategy for both experimental and in-silico data?}

^{3) data-driven multiscale modelling techniques: Can we develop fast and generalizable multiscale models of polymers using experimental and in-silico data?}