The dynamical behavior of molecular systems of relevance to chemistry, biophysics and materials science can be numerically simulated using deterministic molecular dynamics algorithms or stochastic algorithms such as Langevin dynamics. Although the systems of interest are composed of large numbers of atoms, collective interactions mean that the long-time evolution is in fact typically dictated by the variations of a small number of collective modes, known as collective variables or reaction coordinates. Intense efforts have recently been invested in automating the definition of collective variables from molecular simulation data using a variety of machine learning techniques. A key mathematical question is to characterize the quality of the dimensionality reduction, for instance by a priori or, even better, a posteriori estimates on the error committed by integrating the dynamics associated with the coarse-grained reduced model. Another important issue is how to incorporate various constraints into the discovery process, such as symmetries (permutation, rotation, translation). This workshop will focus on recent advances in the data driven learning and validation of collective modes and their applications in coarse-grained simulations and enhanced sampling.