Supervised Matrix Factorization: Local Landscape Analysis and Applications
Authors: Joowon Lee, Hanbaek Lyu, Weixin Yao
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also propose a novel GPU-friendly neural implementation of the BCD algorithm and validate our theoretical findings through numerical experiments. |
| Researcher Affiliation | Academia | 1Department of Mathematics, University of Wisconsin Madison, WI, USA 2Department of Statistics, University of Wisconsin Madison, WI, USA 3Department of Statistics, University of California, Riverside, CA, USA. |
| Pseudocode | Yes | Algorithm 1 BCD algorithm for SMF-W |
| Open Source Code | No | The paper describes a 'GPU-friendly neural implementation' but does not provide a specific link or explicit statement about the public availability of its source code. |
| Open Datasets | Yes | The first dataset is generated from the MNIST database (Le Cun & Cortes, 2010) and the second dataset is a text dataset named Employment Scam Aegean Dataset (Laboratory of Information and Communication Systems, 2016). We apply the proposed methods to two datasets from the Curated Microarray Database (Cu Mi Da) (Feltes et al., 2019). |
| Dataset Splits | No | For the semi-synthetic MNIST dataset, the paper describes the generation of data and noise, but it does not specify explicit training, validation, or test splits for an existing dataset. For the job postings dataset, it mentions its size and class imbalance but no specific split information. |
| Hardware Specification | Yes | All numerical experiments were performed on a workstation with Xeon Gold 6248R @ 3.00GHz CPU, 512GM of RAM, and two RTX A6000 GPUs. |
| Software Dependencies | No | The paper mentions 'TF-IDF normalization (Pedregosa et al., 2011)', implying the use of scikit-learn, but it does not provide specific version numbers for any software components or libraries used. |
| Experiment Setup | Yes | We used Algorithms 1 and 2 with r = 20 for both datasets. The feature matrix Xdata Rp n is then generated by adding an independent Gaussian noise εj N(0, σ2Ip) to the jth column of X0 for j = 1, . . . , n, with σ = 0.5. |