It is crucial to the long-term success of AI systems that they be deployed in a fair and unbiased manner. Think about the challenge of determining whether or not a given image contains a face. In this article, we will look into one recently published method for combating algorithmic bias. To train a de-biased model, we'll construct a facial identification system that discovers the latent variables underlying face picture datasets, and then uses this knowledge to adaptively re-sample the training data. In this experiment, we will use three different data sets. We require both a positive dataset (containing instances of faces) and a negative dataset (containing examples of non-face objects) to train our facial detection models. We'll use these samples to teach our models to recognize faces and other objects in photos. Finally, a test dataset of facial photographs is required. The test dataset we utilize should have representative samples from all of the relevant demographics or traits of interest, as we are concerned about the possibility of bias in the learning models we employ. We'll be thinking about gender and skin tone in this experiment.