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..
Multi-View Oriented GPLVM: Expressiveness and Efficiency
Authors: Zi Yang, Ying Li, Zhidi Lin, Michael Minyi Zhang, Pablo M. Olmos
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical evaluations across a diverse range of multi-view datasets demonstrate that our proposed method consistently outperforms state-of-the-art models in learning meaningful latent representations. |
| Researcher Affiliation | Academia | Zi Yang School of Computing and Data Science University of Hong Kong EMAIL Ying Li School of Computing and Data Science University of Hong Kong EMAIL Zhidi Lin School of Computing and Data Science University of Hong Kong EMAIL Michael Minyi Zhang School of Computing and Data Science University of Hong Kong EMAIL Pablo M. Olmos Department of Signal Theory and Communications Universidad Carlos III de Madrid EMAIL |
| Pseudocode | Yes | Algorithm 1: NG-MVLVM: Next-Gen Multi-View Latent Variable Model |
| Open Source Code | Yes | Code is publicly available at https://github.com/ziyang18/NG-MVLVM. |
| Open Datasets | Yes | We validate our model on a range of cross-domain multi-view datasets, including synthetic, image, text, and wireless communication data. The results show that our model consistently outperforms various state-of-the-art (SOTA) MV-VAEs, MV-GPLVMs, and multi-view extensions of SOTA GPLVMs in terms of generating informative unified latent representations. ...using a high-fidelity channel simulator QUAsi Deterministic Rad Io channel Gener Ator (QUADRIGA). ...following the setting of Wu and Goodman [11], we construct multi-view settings by pairing each single-view dataset images (MNIST, YALE, CIFAR), text (NEWSGROUPS), and structured data (BRIDGES)7 with its corresponding label as an additional view. |
| Dataset Splits | Yes | After inferring the unified latent variable X, we perform five-fold cross-validation using two types of classifiers: K-nearest neighbor (KNN) and support vector machine (SVM). |
| Hardware Specification | Yes | All experiments were conducted on a cloud server equipped with 2 NVIDIA Tesla V100 GPUs (16GB memory each), a 10-core Intel Xeon Platinum 81xx series CPU, 64GB RAM, and 200GB storage. |
| Software Dependencies | No | The paper does not explicitly list software dependencies with specific version numbers. It mentions 'modern optimization techniques, such as Adam [38]', and 'scikit-learn library [49]' for PCA/LDA/ISOMAP, but without specific versions for PyTorch, Python, or other core libraries used in their own implementation. |
| Experiment Setup | Yes | The default hyperparameter settings are summarized in Table 7. Table 7: Default Hyperparameter Settings. PARAMETER DESCRIPTION VALUES NG-SM Kernel Setup # Mixture densities (Q) 2 Dim. of random feature (L/2) 50 Dim. of latent space (D) 2 Optimizer Setup (Adam) Learning rate 0.01 Beta (0.9, 0.99) # Iterations 10000 |