Random Features for Grassmannian Kernels
The Grassmannian manifold G(k, n) serves as a fundamental tool in signal processing, computer vision, and machine learning, where problems often involve classifying, clustering, or comparing subspaces. In this work, we propose a sketching-based approach to approximate Grassmannian kernels using random projections. We introduce three variations of kernel approximation, including two that rely on binarised sketches, offering substantial memory gains. We establish theoretical properties of our method in the special case of G(1, n) and extend it to general G(k, n). Experimental validation demonstrates that our sketched kernels closely match the performance of standard Grassmannian kernels while avoiding the need to compute or store the full kernel matrix. Our approach enables scalable Grassmannian-based methods for large-scale applications in machine learning and pattern recognition.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Grassmannian Packings in Neural Networks: Learning with Maximal Subspace Packings for Diversity and Anti-Sparsity
Kernel sparsity ("dying ReLUs") and lack of diversity are commonly observed in CNN kernels, which decreases model capacity. Drawing inspiration from information theory and wireless communications, we demonstrate the inte…
DiversityExpanding the Family of Grassmannian Kernels: An Embedding Perspective
Modeling videos and image-sets as linear subspaces has proven beneficial for many visual recognition tasks. However, it also incurs challenges arising from the fact that linear subspaces do not obey Euclidean geometry, b…
ClusteringA Study of Neural Collapse Phenomenon: Grassmannian Frame, Symmetry and Generalization
In this paper, we extend original Neural Collapse Phenomenon by proving Generalized Neural Collapse hypothesis. We obtain Grassmannian Frame structure from the optimization and generalization of classification. This stru…
Random Features for Compositional Kernels
We describe and analyze a simple random feature scheme (RFS) from prescribed compositional kernels. The compositional kernels we use are inspired by the structure of convolutional neural networks and kernels. The resulti…
DiversityRandom Gegenbauer Features for Scalable Kernel Methods
We propose efficient random features for approximating a new and rich class of kernel functions that we refer to as Generalized Zonal Kernels (GZK). Our proposed GZK family, generalizes the zonal kernels (i.e., dot-produ…