In this work we demonstrate how to leverage our recent iterative deep learning-all atom molecular dynamics (MD) technique 'Reweighted autoencoded variational Bayes for enhanced sampling (RAVE)' (Ribeiro, Bravo, Wang, Tiwary, J. Chem. Phys. 149, 072301 (2018)) for sampling protein-ligand unbinding mechanisms and calculating absolute binding affinities when plagued with difficult to sample rare events. RAVE iterates between rounds of MD and deep learning, and unlike other enhanced sampling methods, it stands out in simultaneously learning both a low-dimensional physically interpretable reaction coordinate (RC) and associated free energy. Here, we introduce a simple but powerful extension to RAVE which allows learning a position-dependent RC expressed as a superposition of piecewise linear RCs valid in different metastable states. With this approach, we retain the original physical interpretability of a RAVE-derived RC while making it applicable to a wider range of complex systems. We demonstrate how in its multi-dimensional form introduced here, RAVE can efficiently simulate the unbinding of the tightly bound benzene-lysozyme (L99A variant) complex, in all atom-precision and with minimal use of human intuition except for the choice of a larger dictionary of order parameters. These simulations had a 100% success rate, and took between 3-50 nanoseconds for a process that takes on an average close to few hundred milliseconds, thereby reflecting a seven order of magnitude acceleration relative to straightforward MD. Furthermore, without any time-dependent biasing, the trajectories display clear back-and-forth movement between various metastable intermediates, demonstrating the reliability of the RC and its probability distribution learnt in RAVE. Our binding free energy is in good agreement with other reported simulation results. We thus believe that RAVE, especially in its multi-dimensional variant introduced here, will be a useful tool for simulating the dissociation process of practical biophysical systems with rare events in an automated manner with minimal use of human intuition.