Disentangled Representations are representations in machine learning where different factors of variation in the data are separated into distinct components. This allows for better interpretability and control over the learned representations, making them useful for tasks like generative modeling and transfer learning.