Granularity Statistical Invariants Learning
Tingting Zhu · 2024
Learning using statistical invariants (LUSI) is a new paradigm in machine learning, which aims to effectively handle classification problems by utilizing weak convergence mechanisms. However, despite the wide applicability demonstrated by LUSI in handling classification tasks, training large-scale datasets poses computational challenges due to the high cost of computing invariant matrices in invariant learning. Although there have been recent improvements in related research, they are still not comprehensive enough. Therefore, this paper extends it to other statistical invariant learning mechanisms in Hilbert space, resulting in a more comprehensive granularity statistical invariant learning model, namely granularity statistical invariant learning (GSIL). Finally, experimental validation demonstrates that GSIL effectively resolves the storage and computation issues of the V -matrix classification method, while also enhancing the generalization capability, particularly exhibiting significant advantages in nonlinear spaces.