PIP (Perturbation-based Iterative Pruning), a novel double-view structured pruning method to optimize LLMs, which combines information from two different views: the unperturbed view and the perturbed view, establishing it as a leading technique for optimizing LLMs in environments with constrained resources.

We propose a spectral clustering method based on data-dependent balls, which only takes \(\mathcal{O}(n)\) time to construct the balls and the number of balls can be fixed. It is thus the first spectral clustering algorithm to our knowledge that constructs a similarity matrix of constant size and an eigendecomposition of constant time complexity. Experiments show that our method has lower memory and time complexity and has better clustering effects than GBSC.

We propose a novel distribution-based spectral clustering. Our method constructs an \(n \times k \) bipartite graph between n data points and k distributions, enabling the eigendecomposition of only a \(k \times k\) matrix and preserving clustering quality at the same time.