Comes from doi.org/10.3389/fcomp.2020.00001

ADR is a vectorization/encoding technique whereby time-series data is segmented, clustered via solf-organizing maps, and the centroids of the clusters are used as the encoding

create a space of molecules of its ocnstructions, and use active learning to search through it.

Create an approximation of the value function \(U_{\phi}\) using Approximate Value Function, and use Policy Gradient to optimize an monte-carlo tree search policy