Co-manifold learning with missing data
Authors: Gal Mishne, Eric Chi, Ronald Coifman
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate that our approach outperforms competing methods in both data visualization and clustering. In Figure 2, we display the embeddings of the different methods for each of the three datasets for both their rows (top) and their columns (bottom), where 50% of the entries have been removed. The top panel of Figure 3 compares clustering the embedding of the cancer patients in lung500 by each method for increasing percentage of missing values in the data, where we averaged over 30 realizations of missing entries. |
| Researcher Affiliation | Academia | 1Department of Mathematics, Yale University, New Haven, CT, USA 2Department of Statistics, North Carolina State University, Raleigh, NC, USA. |
| Pseudocode | Yes | Algorithm 1 CO-CLUSTER-MISSING(PΘ(X), γr, γc) ... Algorithm 2 Co-manifold learning on an Incomplete Data Matrix |
| Open Source Code | No | The paper does not provide any statement or link regarding open-source code for the described methodology. |
| Open Datasets | Yes | lung500 A real-world dataset composed of 56 lung cancer patients and their gene expression (Lee et al., 2010). |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset split percentages or counts for its experiments. It focuses on varying percentages of missing values in the data. |
| Hardware Specification | No | The paper does not provide any specific hardware details used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers used for replication. |
| Experiment Setup | Yes | For concreteness, in the rest of this paper, we use the following function Ω = 1/ζ + ϵdζ, where ϵ is a small positive number, e.g. 10^-12. ... we set α = 1/2 to favor local over global structure in our simulations. ... We start with small values of γr = 2l0 and γc = 2k0, where l0, k0 < 0. ... The matrix U(0) is initialized to be the mean of all non-missing values. |