Efficient Label Propagation
Authors: Yasuhiro Fujiwara, Go Irie
ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate the significant superiority of our algorithm over existing label propagation methods. We performed experiments to compare the proposed approach to the optimal solution and the power method in terms of efficiency and effectiveness. The experiments used the following standard datasets. Reuters-21578, COIL-100. |
| Researcher Affiliation | Industry | Yasuhiro Fujiwara FUJIWARA.YASUHIRO@LAB.NTT.CO.JP NTT Software Innovation Center, 3-9-11 Midori-cho Musashino-shi, Tokyo, Japan Go Irie IRIE.GO@LAB.NTT.CO.JP NTT Media Intelligence Laboratories, 1-1 Hikarinooka Yokosuka-shi, Kanagawa, Japan |
| Pseudocode | Yes | Algorithm 1 Proposed algorithm |
| Open Source Code | No | No, the paper does not include any statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | Reuters-21578 1: This dataset contains documents released by the Reuters newswire. Documents with multiple category labels were discarded. As a result, it contained 8, 293 documents of 65 categories. tf-idf was used as the document feature; it has 18, 933 dimensions. 1http://www.daviddlewis.com/resources/testcollections/reuters21578/ COIL-100 2: This dataset contains images of 100 objects; the number of object labels is 100. Images of the objects were taken at pose intervals of 5 degrees; 72 poses per object resulting in 7, 200 images. We resized all images to 32 32 and used RGB pixel values as the feature vector, resulting in 3, 048 dimensions. 2http://www.cs.columbia.edu/CAVE/software/softlib/coil-100.php |
| Dataset Splits | No | No, the paper mentions that “10 data points in each category/object were initially labeled,” but it does not provide specific train/validation/test split percentages or counts for the entire dataset needed for full reproducibility of data partitioning in a conventional sense. |
| Hardware Specification | Yes | All experiments were conducted on a Linux 2.70 GHz Intel Xeon sever. |
| Software Dependencies | No | No, the paper mentions general software context but does not provide specific names of libraries, solvers, or packages with their version numbers. |
| Experiment Setup | Yes | Following previous papers (Zhou et al., 2003; Xu et al., 2011), we set α = 0.99 and stop iterating the power method when the residual drops below 10^-4. In the experiments, 10 data points in each category/object were initially labeled. 100 nearest neighbors were used to construct each graph. We set the parameter λ = 10 on sparse L1 graph. |