A training technique that simulates viewing images from different angles and perspectives to teach the model to recognize the same features under geometric transformations.