The paper "Taskonomy: Disentangling Task Transfer Learning" by Amir R. Zamir et al. explores the relationship between various visual tasks and proposes a computational approach to model this structure. The authors aim to reduce the need for labeled data by leveraging the relationships among tasks, which can lead to more efficient and effective transfer learning. They use a dictionary of 26 2D, 2.5D, 3D, and semantic tasks to find transfer learning dependencies in a latent space. The resulting computational taxonomic map, or taskonomy, helps identify non-trivial relationships between tasks and reduces the number of labeled data points required to solve a set of tasks. The paper includes an interactive taxonomy solver, a dataset, and code available at http://taskonomy.vision/. The method is evaluated on common datasets like ImageNet and Places, showing significant reductions in the number of labeled data points needed while maintaining performance. The authors also demonstrate the generalization of the computed taxonomies to novel tasks and evaluate the significance of the discovered task space structure.The paper "Taskonomy: Disentangling Task Transfer Learning" by Amir R. Zamir et al. explores the relationship between various visual tasks and proposes a computational approach to model this structure. The authors aim to reduce the need for labeled data by leveraging the relationships among tasks, which can lead to more efficient and effective transfer learning. They use a dictionary of 26 2D, 2.5D, 3D, and semantic tasks to find transfer learning dependencies in a latent space. The resulting computational taxonomic map, or taskonomy, helps identify non-trivial relationships between tasks and reduces the number of labeled data points required to solve a set of tasks. The paper includes an interactive taxonomy solver, a dataset, and code available at http://taskonomy.vision/. The method is evaluated on common datasets like ImageNet and Places, showing significant reductions in the number of labeled data points needed while maintaining performance. The authors also demonstrate the generalization of the computed taxonomies to novel tasks and evaluate the significance of the discovered task space structure.