The rate of perceptual learning at a fixed accuracy threshold is improved by feedback and by mixture with easier trials