Shared Generative Latent Representation Learning for Multi-View Clustering
Authors: Ming Yin, Weitian Huang, Junbin Gao6688-6695
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The extensive experimental results on several datasets with different scales demonstrate that the proposed method outperforms the state-of-the-art methods under a range of performance criteria. |
| Researcher Affiliation | Academia | Ming Yin, 1 Weitian Huang,1 Junbin Gao2 1School of Automation, Guangdong University of Technology, Guangzhou 510006, China. 2The University of Sydney Business School, The University of Sydney, Camperdown, NSW 2006, Australia. |
| Pseudocode | No | The paper describes the method and architecture but does not include pseudocode or a clearly labeled algorithm block. |
| Open Source Code | No | The paper does not provide any concrete access to source code for the described methodology. |
| Open Datasets | Yes | UCI digits2 consists of features of handwritten digits of 0 to 9 extracted from UCI machine learning repository. It contains 2000 data points with 200 samples for each digit. These digits are represented by six types of features, including pixel averages in 2 3 windows (PIX) of dimension 240, Fourier coefficients of the character shapesdimension 76, profile correlations (FAC) of dimension 216, Zernike moments (ZER) of dimension 47, Karhunen Loeve coefficients (KAR) of dimension 64 and morphological features (MOR) of dimension 6. (https://archive.ics.uci.edu/ml/datasets/Multiple+Features). Caltech 101 is an object recognition dataset (Li, Fergus, and Perona 2004) containing 8677 images of 101 categories. |
| Dataset Splits | No | The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning. It mentions "After pre-training, the network g( ) is adopted to project input data points into the latent representation z, and then we perform K-means to z to obtain K initial centroids of GMM μc(c {1, , K})." |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions "Adam optimizer" but does not provide specific software names with version numbers for libraries or frameworks used. |
| Experiment Setup | Yes | In our experiments, the fully connected network and same architecture settings as DEC (Xie, Girshick, and Farhadi 2016) are used. More specifically, the architectures of g(x(v); φ(v)) and f(z; θ(v)) are dv-500-500-200-10 and 10-200-500-500-dv, respectively, where dv is input dimensionality of each view. We use Adam optimizer (Kingma and Ba 2015) to maximize the objective function, and set the learning rate to be 0.0001 with a decay of 0.9 for every 10 epochs. Besides, the weights w of Eqs. (6) and (7) are initialized to 1 m for each view and the parameter of GMM πk is initialized to 1 K . |