Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
ComRank: Ranking Loss for Multi-Label Complementary Label Learning
Authors: Jing-Yi Zhu, Yi Gao, Miao Xu, Min-Ling Zhang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate the effectiveness of our method in MLCLL tasks. The code is available at https://github.com/Jelly Jam Zhu/Com Rank. ... Outstanding experimental results demonstrate the effectiveness of our method. ... Section 6 Experiments |
| Researcher Affiliation | Academia | Jing-Yi Zhu1,2, Yi Gao1,2 , Miao Xu3, Min-Ling Zhang1,2 1School of Computer Science and Engineering, Southeast University, Nanjing 210096, China 2Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education, China 3The University of Queensland, Australia EMAIL, {miao.xu}@uq.edu.au |
| Pseudocode | Yes | Algorithm 1 Com Rank Algorithm Input: D: the complementary-label training set {(xi, yi)}n i=1 θ: the initial parameters of classifier g T: the number of epochs A: an external stochastic optimization algorithm Output: g: learned multi-label classifier Training Routine 1: for t = 1 to T do 2: Let L be the risk, 3: L = 1 n Pn i=1{ LCR(g(xi), yi)} 4: Set gradients θL 5: Update θ by A with θL 6: end for |
| Open Source Code | Yes | Experiments demonstrate the effectiveness of our method in MLCLL tasks. The code is available at https://github.com/Jelly Jam Zhu/Com Rank. |
| Open Datasets | Yes | To fully verify the effectiveness of Com Rank, we select seven multi-label datasets for experiments2. ... 2Publicly available at https://mulan.sourceforge.net/datasets-mlc.html. |
| Dataset Splits | Yes | For statistical analysis, we employ ten-fold cross-validation, where the dataset is randomly divided into ten subsets. |
| Hardware Specification | Yes | Our experiments are conducted using Py Torch [Paszke et al., 2019] and implement on an NVIDIA TITAN RTX. |
| Software Dependencies | No | Our experiments are conducted using Py Torch [Paszke et al., 2019] and implement on an NVIDIA TITAN RTX. The paper mentions PyTorch but does not provide a specific version number. The citation [Paszke et al., 2019] refers to the original PyTorch paper, not necessarily the version used for the experiments. |
| Experiment Setup | Yes | The weight decay was set to 1e 3, and the learning rate was selected from {1e 3, 1e 2, 1e 1}. It is multiplied by 0.1 when the iteration count reaches 100 and 150 [Wang et al., 2021]. The training epochs for all datasets are 200. These settings were kept consistent across all methods. |