Efficient non‐linear model reduction via a least‐squares Petrov–Galerkin projection and compressive tensor approximations | Synapse