Fast Updating Truncated SVD for Representation Learning with Sparse Matrices
Authors: Haoran Deng, Yang Yang, Jiahe Li, Cheng Chen, Weihao Jiang, Shiliang Pu
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical experiments demonstrate the efficiency of our method is improved by an order of magnitude compared to previous methods. Remarkably, this improvement is achieved while maintaining a comparable precision to existing approaches. Codes are available1. 4 NUMERICAL EXPERIMENT In this section, we conduct experimental evaluations of the update process for the truncated SVD on sparse matrices. |
| Researcher Affiliation | Collaboration | Haoran Deng1, Yang Yang1 , Jiahe Li1, Cheng Chen2, Weihao Jiang2, Shiliang Pu2 1Zhejiang University, 2Hikvision Research Institute |
| Pseudocode | Yes | Algorithm 3: Add columns |
| Open Source Code | Yes | Codes are available1. 1https://github.com/zjunet/Inc SVD. We released the implement of all the algorithms involved in the experiment as a python package. The code is available at: https://github.com/zjunet/Inc SVD. |
| Open Datasets | Yes | Datasets. The link prediction experiments are conducted on three publicly available graph datasets, namely Slashdot (Leskovec et al., 2009) with 82, 168 nodes and 870, 161 edges, Flickr (Tang & Liu, 2009) with 80, 513 nodes and 11, 799, 764 edges, and Epinions (Richardson et al., 2003) with 75, 879 nodes and 1, 017, 674 edges. ... For the collaborative filtering task, we use data from Movie Lens (Harper & Konstan, 2015). |
| Dataset Splits | No | For the link prediction, we create the training set G by randomly removing 30% of the edges from the original graph G. ... For the collaborative filtering task, values in the matrix are initially divided into the training and testing set with a ratio of 8 : 2. No explicit mention of a validation set. |
| Hardware Specification | No | The paper does not explicitly state the specific hardware used (e.g., GPU/CPU models, memory specifications) for running its experiments. |
| Software Dependencies | No | The paper mentions a 'python package' for the implementation but does not specify particular software dependencies with version numbers (e.g., specific library versions like PyTorch 1.9 or TensorFlow 2.x). |
| Experiment Setup | Yes | Throughout the experiments, we set l, the spatial dimension of the approximation, to 10 based on previous settings (Vecharynski & Saad, 2014). The required number of iterations for the RPI, denoted by t, is set to 3. |