We numerically investigate the aging dynamics of a monatomic Lennard-Jones glass, focusing on the topology of the potential energy landscape which, to this aim, has been partitioned in basins of attraction of stationary points (saddles and minima). The analysis of the stationary points visited during the aging dynamics shows the existence of two distinct regimes: (i) at short times the system visits basins of saddles whose energies and orders decrease with t; (ii) at long times the system mainly lies in basins pertaining to minima of slowly decreasing energy. The long time dynamics can be represented by a simple random walk on a network of minima with a jump probability proportional to the inverse of the waiting time.