Many challenges remain in applying machine learning to domains where obtaining massive annotated data is difficult. We discuss a number of approaches that aim to reduce supervision load for learning algorithms in the visual and geometric domains by leveraging correlations among data, among representations, and among learning tasks -- what we call joint learning. The basic notion is that inference problems do not occur in isolation but rather in a social context that can be exploited to provide self-supervision by enforcing consistency among them, thus improving performance and increasing sample efficiency. An example is shape co-segmentation, where we can use structural correlations between related shapes to regularize the segmentation of any particular shape. Another is the use of cross-task consistency constraints, as in the case of inferring depth and normals from an image, which are obviously related. Even at the level of representations, joint learning can avoid blind-spots of any one individual representation and better adapt to data particularities – just as we get with multiple 2D views of a 3D object. The talk will present a number of examples of joint learning, including the above as well as 3D object detection and pose estimation.
