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..
LLM-DAMVC: A Large Language Model Assisted Dynamic Agent for Multi-View Clustering
Authors: Qianqian Wang, Qianqian Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results validate the effectiveness and superiority of the proposed method. |
| Researcher Affiliation | Academia | 1School of Telecommunications Engineering, Xidian University, Xi an, China EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology using mathematical formulations and descriptive text, but does not include any explicitly labeled pseudocode or algorithm blocks. For instance, the training process is described through objective functions and component interactions rather than a step-by-step algorithm. |
| Open Source Code | Yes | We plan to make the code repository publicly available via Git Hub after the paper is published, with detailed usage instructions and reproduction guidelines.The supplementary materials include core code. |
| Open Datasets | Yes | We evaluate our method on six benchmark multi-view datasets to validate its effectiveness across diverse data types: NUS-WIDE (NUS) (37), an image dataset containing 6,251 samples described by 5 visual feature views; BDGP (38), a Drosophila image dataset with 2,500 samples and 2 views (visual and textual); Handwritten (39), a digit recognition dataset comprising 2,000 samples represented by 6 heterogeneous feature views; MNIST-USPS (38), a cross-domain digit benchmark with 5,000 samples from two complementary image datasets; Reuters (40), a multilingual news corpus consisting of 1,800 short articles and their associated topics, represented by 5 language-specific views; and CCV (40), a consumer video dataset with 6,773 samples and 3 deep feature views. |
| Dataset Splits | No | The paper mentions several benchmark multi-view datasets used for evaluation but does not explicitly specify the training, validation, or test splits for any of these datasets within the main text. |
| Hardware Specification | Yes | All experiments are implemented in Py Torch and conducted on an NVIDIA Ge Force RTX 4090 GPU. |
| Software Dependencies | No | The paper states, 'All experiments are implemented in Py Torch,' but does not provide a specific version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | We analyze the sensitivity of our model to the hyperparameters λ, β, and γ that weight the contrastive loss (Lcont), clustering loss (Lclus), and alignment loss (Lalign). The model is robust to the choice of λ: performance remains stable over a wide range of values, which shows that the contrastive loss consistently fulfills its role of pulling similar samples together and pushing dissimilar ones apart, and this function is reliably effective without requiring precise tuning. In contrast, for the parameters γ and β, there exist performance peaks only at specific values and drops when deviating, demonstrating that the alignment loss and clustering loss must be carefully balanced to work properly. The alignment loss is responsible for establishing cross-view consistency by aligning representations from different views into a shared space, and the clustering loss directly optimizes the cluster structure by encouraging samples to form coherent groups. By comprehensively analyzing the three figures, we found that high performance is achieved when all three losses are present. Removing or severely weakening any one of them leads to degradation, which confirms that each loss performs a distinct and necessary function: contrastive learning handles fine-grained sample discrimination, alignment ensures cross-view agreement, and clustering enforces global semantic grouping. |