Singular Subspace Perturbation Bounds via Rectangular Random Matrix Diffusions

Authors: Peiyao Lai, Oren Mangoubi

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present numerical simulations that illustrate the theoretical results in Theorem 2.2, and investigate the extent to which the bounds in Theorem 2.2 are tight.
Researcher Affiliation Academia Peiyao Lai Worcester Polytechnic Institute Worcester, MA, USA Oren Mangoubi Worcester Polytechnic Institute Worcester, MA, USA
Pseudocode No No explicit pseudocode or algorithm blocks are provided in the paper.
Open Source Code No The paper does not contain any explicit statements about releasing source code or links to code repositories.
Open Datasets No In this set of simulations, we compute the squared Frobenius error for the rank-k covariance approximation problem... In the following simulations we choose the input data matrix to be a synthetic data matrix with linearly decaying spectral profile spectral profile σi = m (d i + 1) for all i [d].
Dataset Splits No The paper uses a synthetic data matrix for numerical simulations, but it does not specify any training, testing, or validation dataset splits.
Hardware Specification No The paper includes a 'Numerical Simulations' section, but it does not specify any hardware details (e.g., GPU/CPU models, memory) used for running these simulations.
Software Dependencies No The paper describes numerical simulations but does not provide any specific software dependencies or version numbers (e.g., programming languages, libraries, frameworks).
Experiment Setup Yes In this set of simulations, we compute the squared Frobenius error for the rank-k covariance approximation problem... We take an input data matrix A, perturb the matrix by iid Gaussian noise (that is, A = A + TG where G has iid N(0, 1) entries), and compute the error for different values of m, d, k. In the following simulations we choose the input data matrix to be a synthetic data matrix with linearly decaying spectral profile spectral profile σi = m (d i + 1) for all i [d]... Here, d = 15, k = 5, T = 1 and the input matrix has spectral profile σi = m (d i + 1) for all i [d].