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..
Can Class-Priors Help Single-Positive Multi-Label Learning?
Authors: Biao Liu, Ning Xu, Jie Wang, Xin Geng
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on ten MLL benchmark datasets demonstrate the effectiveness and superiority of our method over existing SPMLL approaches. |
| Researcher Affiliation | Academia | Biao Liu, Ning Xu*, Jie Wang, Xin Geng School of Computer Science and Engineering, Southeast University, Nanjing 210096, China Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China *Corresponding author: EMAIL |
| Pseudocode | Yes | Algorithm 1 CRISP Algorithm Input: The SPMLL training set D = {(xi, γi)|1 i n}, the multi-label classifier f, the number of epoch T, hyperparameters 0 δ, τ 1; 1: for t = 1 to T do 2: for j = 1 to c do 3: Extract the positive-labeled samples set SLj = {xi : lj i = 1, 1 i n}. 4: Estimate ˆqj(z) = 1 n Pn i=1 1(f j(xi) z) and ˆqp j (z) = 1 np j P x SLj 1(f j(x) z) for all 5: Estimate the class-prior of j-th label by ˆπj = ˆqj(ˆz) ˆqp j (ˆz) with the threshold induced by Eq. (2). 6: end for 7: Update the model f by forward computation and back-propagation by Eq. (7) using the estimated class-priors. 8: end for Output: The predictive model f. |
| Open Source Code | No | The code is currently proprietary and not publicly available. However, we have provided detailed information necessary for replicating the experiments. |
| Open Datasets | Yes | Datasets. In the experimental section, our proposed method is evaluated on four large-scale multi-label image classification (MLIC) datasets and six widely-used multi-label learning (MLL) datasets. The four MLIC datasets include PSACAL VOC 2021 (VOC) [8], MS-COCO 2014 (COCO) [20], NUS-WIDE (NUS) [4], and CUB-200 2011 (CUB) [29]; the MLL datasets cover a wide range of scenarios with heterogeneous multi-label characteristics. |
| Dataset Splits | Yes | For each MLIC dataset, 20% of the training set is withheld for validation. Each MLL dataset is partitioned into train/validation/test sets at a ratio of 80%/10%/10%. |
| Hardware Specification | Yes | The code implementation is based on Py Torch, and the experiments are conducted on Ge Force RTX 3090 GPUs. |
| Software Dependencies | No | The code implementation is based on Py Torch, and the experiments are conducted on Ge Force RTX 3090 GPUs. |
| Experiment Setup | Yes | The batch size is selected from {8, 16} and the number of epochs is set to 10. The learning rate and weight decay are selected from {10 2, 10 3, 10 4, 10 5} with a validation set. The hyperparameters δ and τ are all fixed as 0.01. All the comparing methods run 5 trials on each datasets. For fairness, we employed Res Net-50 as the backbone for all comparing methods. |