Our constrained optimization approach effectively mitigates the disparate impact of pruning in sparse models.