Researchers have developed a new image translation model that can reduce biases in data used for training artificial intelligence (AI) models. These biases can occur when diverse sources of photos are used, leading to inconsistencies in image analysis. The new model addresses this by constructing a dataset using texture debiasing and training the model on that dataset. The model outperforms existing methods and can be applied to various domains, such as medicine and self-driving cars, to enhance the robustness of AI models.
Study creates AI model for effectively reducing biases in dataset – Times of India
