Neural optimal control . Neural optimal control (A) is a way of optimally steering a dynamical system (here with equation of motion f) by learning an adaptive control system (with equation of motion g) from data. In recent joint work with Mathieu Chalivdal, I showed how this method could be used to build networks which could reconstruct (B) a recently viewed high-dimensional pattern from memory far better (C) (blue loss curve) than competing models.
Articles in preparation
Selected Articles