Which technique should an AI practitioner use to address bias in a model generating images of humans in various professions, when specific attributes in the input data are causing biased image generation?