LazySVD: Even Faster SVD Decomposition Yet Without Agonizing Pain
Authors: Zeyuan Allen-Zhu, Yuanzhi Li
NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the practicality of our Lazy SVD framework, and compare it to block power method or block Krylov method. We use datasets SNAP/amazon0302, SNAP/email-enron, and news20 that were also used by Musco and Musco [19], as well as an additional but famous dataset RCV1. The four datasets give rise sparse matrices of dimensions 257570 262111, 35600 16507, 11269 53975, and 20242 47236 respectively. Our Results. We study four datasets each with k = 10, 20, 30 and with the four performance metrics, totaling 48 plots. Due to space limitation, we only select six representative plots out of 48 and include them in Figure 1. Lazy SVD outperforms its three competitors almost universally. |
| Researcher Affiliation | Academia | Zeyuan Allen-Zhu zeyuan@csail.mit.edu Institute for Advanced Study & Princeton University Yuanzhi Li yuanzhil@cs.princeton.edu Princeton University |
| Pseudocode | Yes | Algorithm 1 Appx PCA(A, M, δ , ε, p) (only for proving our theoretical results; for practitioners, feel free to use your favorite 1-PCA algorithm such as Lanczos to replace Appx PCA.) Algorithm 2 Lazy SVD(A, M, k, δ , εpca, p) |
| Open Source Code | No | The paper states 'We programmed the four algorithms using the same programming language with the same sparse-matrix implementation' but does not provide any link or explicit statement about releasing the source code for their methodology. |
| Open Datasets | Yes | We use datasets SNAP/amazon0302, SNAP/email-enron, and news20 that were also used by Musco and Musco [19], as well as an additional but famous dataset RCV1. The first two can be found on the SNAP website [16] and the last two can be found on the Lib SVM website [11]. |
| Dataset Splits | No | The paper mentions datasets used for experiments but does not provide specific details on training, validation, and test splits (e.g., percentages or sample counts). |
| Hardware Specification | Yes | We tested them single-threaded on the same Intel i7-3770 3.40GHz personal computer. |
| Software Dependencies | No | The paper states 'We programmed the four algorithms using the same programming language with the same sparse-matrix implementation' but does not provide specific software names with version numbers (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | No | The paper describes the implemented algorithms and general running time comparison, but it does not provide specific experimental setup details such as concrete hyperparameter values, optimizer settings, or detailed training configurations in the main text. |