Label Distribution Learning Machine

Authors: Jing Wang, Xin Geng

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results validate the better classification performance of LDLM.
Researcher Affiliation Academia 1School of Computer Science and Engineering, Southeast University, Nanjing, China 2Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education. Correspondence to: Xin Geng <xgeng@seu.edu.cn>.
Pseudocode Yes The details of the algorithm are presented in Algorithm 1
Open Source Code No The paper only provides links to open-source code for baseline methods (EDL-LRL and LDLFs) and not for their own proposed methodology (LDLM).
Open Datasets Yes The first 15 datasets are collected by Geng (2016), where the first ten (from Alpha to Spoem) are from the clustering analysis of genome-wide expression in Yeast Saccharomyces cerevisiae (Eisen et al., 1998), the Scene is a multi-label image dataset whose label distributions are transformed from rankings (Geng & Luo, 2014), the Gene is obtained from the research on the relation between gene and diseases (Yu et al., 2012), the Movie is collected from user ratings on movies (Geng & Hou, 2015), and the SJAFFE and SBU 3DFE are collected from JAFFE (Lyons et al., 1998) and BU 3DFE (Yin et al., 2006), respectively. The M2B (Nguyen et al., 2012) and SCUT-FBP (Xie et al., 2015) are about facial beauty perception, which are pre-processed as (Ren & Geng, 2017).
Dataset Splits Yes We tune the parameters of each method by ten-fold cross-validation.
Hardware Specification Yes Moreover, we implement LDLM in Python and carry out the experiments on a Linux server with a 2.70GHz CPU and 62GB memory.
Software Dependencies No The paper states "we implement LDLM in Python" but does not provide specific version numbers for Python or any other key software libraries or dependencies.
Experiment Setup Yes For LDLM, λ1 = 0.001, λ2 and λ3 are tuned from the candidate set {10 3, , 1}, and ρ = 0.01. We tune the parameters of each method by ten-fold cross-validation.