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..
Bit-swapping Oriented Twin-memory Multi-view Clustering in Lifelong Incomplete Scenarios
Authors: Shengju Yu, Pei Zhang, Siwei Wang, Suyuan Liu, Xinhang Wan, Zhibin Dong, Tiejun Li, Xinwang Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments organized on publicly-available benchmark datasets with varying missing percentages confirm the superior effectiveness of our BSTM. Extensive contrast experiments with diverse missing data ratios on several public datasets. |
| Researcher Affiliation | Academia | 1National University of Defense Technology, 2Intelligent Game and Decision Lab EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 Bit-swapping Oriented Twin-memory Lifelong Incomplete MVC (BSTM) Input: Incomplete multi-view data Dt with index vector wt, hyper-parameters λ and β; Initialize: G1, e B1, e Q1, b S1; Output: Performing spectral partitioning on e Qt to produce clustering results; 1: for t = 1 to v do 2: if t == 1 then 3: Setting λ and β as 0. 4: end if 5: l = 0; 6: repeat 7: l = l + 1; 8: Updating the guidance variable Gt via (8); 9: Updating the unified anchor variable e Bt via (10); 10: Updating the shared bipartite graph e Qt via (15); 11: if t > 1 then 12: Updating the swapping variable b St 1 via (18); 13: end if 14: until l > 1 and floss(l 1) floss(l) floss(l 1) ϵ; 15: end for |
| Open Source Code | No | Answer: [No] Justification: The used benchmark datasets are public. Code will be released under license in the final version. |
| Open Datasets | Yes | Extensive experiments organized on publicly-available benchmark datasets with varying missing percentages confirm the superior effectiveness of our BSTM. We conduct experiments on 8 public datasets, with their key characteristics summarized in Table 1. The used benchmark datasets are public. |
| Dataset Splits | No | The paper specifies different 'missing ratios' for the data in the experiments (0.1, 0.4, and 0.7), but it does not provide explicit training, validation, or test dataset splits. The description of experiments focuses on varying data incompleteness rather than standard data partitioning methodologies for reproducibility. |
| Hardware Specification | Yes | Answer: [Yes] Justification: We conduct experiments on Window 10 with 64GB memory and Intel i7 CPU. Please see Fig. 2 for the execution time. |
| Software Dependencies | No | The paper mentions 'Window 10' as the operating system for experiments but does not provide specific version numbers for any programming languages, libraries, frameworks, or solvers used in the implementation of the methodology. |
| Experiment Setup | Yes | In experiments, we initialize the variables G1 and b S1 using a random orthogonal matrix and an identify matrix, respectively. For e B1, we firstly construct a unified space by minimizing the reconstruction error, and then employ k-means within this space to generate initial anchors serving as the assignment of e B1. For e Q1, we utilize one-hot vectors to randomly assign its columns, guaranteeing that the sum of its each column equals to 1. For hyper-parameters λ and β, we fine-tune them in [10 1, 100, 101, 102, 103] respectively. For the stopping error ϵ, we set it as 10 4. Then, we execute 20 times and summarize the average clustering results. |