SMS dit: Dynamic Depth Learning in Stacked AutoEncoders