Learning Low-Dimensional Metrics

Authors: Blake Mason, Lalit Jain, Robert Nowak

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 3 Experiments To validate our complexity and recovery guarantees, we ran the following simulations. We generate x1, , xn iid N(0, 1 p I), with n = 200, and K = p p d UU T for a random orthogonal matrix U 2 Rp d with unit norm columns. In Figure 2a, K has d nonzero rows/columns. In Figure 2b, K is a dense rank-d matrix. We compare the performance of nuclear norm and 1,2 regularization in each setting against an unconstrained baseline where we only enforce that K be psd. Given a fixed number of samples, each method is compared in terms of the relative excess risk, R(c K) R(K ) R(K ) , and the relative squared recovery error, kc K K k2 F , averaged over 20 trials. The y-axes of both plots have been trimmed for readability.
Researcher Affiliation Academia Lalit Jain University of Michigan Ann Arbor, MI 48109 lalitj@umich.edu Blake Mason University of Wisconsin Madison, WI 53706 bmason3@wisc.edu Robert Nowak University of Wisconsin Madison, WI 53706 rdnowak@umich.edu
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not contain any explicit statement about providing source code or a link to a code repository for the described methodology.
Open Datasets No The paper states 'We generate x1, , xn iid N(0, 1 p I), with n = 200' for their simulations, indicating a self-generated dataset for which no public access information (link, DOI, citation) is provided.
Dataset Splits No The paper describes simulations but does not provide specific details regarding training, validation, or test dataset splits, percentages, or sample counts.
Hardware Specification No The paper describes running simulations but does not provide specific hardware details (e.g., GPU/CPU models, memory, or processor types) used for the experiments.
Software Dependencies No The paper does not provide specific software dependency details, such as library or solver names with version numbers.
Experiment Setup Yes We generate x1, , xn iid N(0, 1 p I), with n = 200, and K = p p d UU T for a random orthogonal matrix U 2 Rp d with unit norm columns. In Figure 2a, K has d nonzero rows/columns. In Figure 2b, K is a dense rank-d matrix. ... averaged over 20 trials. ... we compute the number of samples averaged over 20 runs to achieve a relative excess risk of less than 0.1 in Figure 3. First, we fix p = 100 and increment d from 1 to 10. Then we fix d = 10 and increment p from 10 to 100 to clearly show the linear dependence of the sample complexity on d and p as demonstrated in Corollary 2.1.1.