Joint Learning Over Visual and Geometric Data
Explore joint learning approaches for reducing supervision load in visual and geometric domains in this Stanford University seminar. Delve into Professor Leonidas Guibas' discussion on leveraging correlations among data, representations, and learning tasks to improve performance and increase sample efficiency. Discover how enforcing consistency among inference problems provides self-supervision in a social context. Examine topics such as information transportation and aggregation, PointNet structure, point cloud object detection, multi-modal learning with point clouds and RGB images, spatiotemporal generation, rigid reconstruction, and reference language games. Gain insights into cutting-edge techniques for addressing challenges in machine learning applications where massive annotated datasets are difficult to obtain.