Multi-View Information-Bottleneck Representation Learning
Authors: Zhibin Wan, Changqing Zhang, Pengfei Zhu, Qinghua Hu10085-10092
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments (including clustering analysis, robustness experiment, and ablation study) on real-world datasets, which empirically show promising generalization ability and robustness compared to state-of-the-arts.In the experiments, we compare our proposed CMIBNets with existing state-of-the-art multi-view representation learning algorithms on real-world multi-view datasets. Specifically, we evaluate the performances on clustering in terms of common metrics, and verify the generalization and robustness of the model through a variety of experiments.We evaluate our model on six multi-view benchmark datasets in the experiments, including: 1) Handwritten1... Table 1: Performance comparison on clustering task. |
| Researcher Affiliation | Academia | Zhibin Wan,1 Changqing Zhang,1,2 Pengfei Zhu,1,2 Qinghua Hu1,2 1College of Intelligence and Computing, Tianjin University, Tianjin, China 2Tianjin Key Lab of Machine Learning, Tianjin, China {wanzhibin, zhangchangqing, zhupengfei, huqinghua}@tju.edu.cn |
| Pseudocode | No | No pseudocode or clearly labeled algorithm block was found in the paper. |
| Open Source Code | No | The paper does not contain any explicit statement offering open-source code for the described methodology or a direct link to a code repository. |
| Open Datasets | Yes | We evaluate our model on six multi-view benchmark datasets in the experiments, including: 1) Handwritten1 contains 2000 examples of numbers 09 with 200 samples per class. These images are represented with two different types of features. 2) ORL2 face dataset includes 40 different categories, and each category has 10 different facial images. The intensity and Gabor are used as different views. 3) COIL-203 consists of 1440 pictures of 20 categories. In the experiments, the intensity and Gabor features are extracted as two different perspectives. 4) Caltech101-74 is a subset of the original Caltech101 image dataset. This subset selected 1,474 images in seven views. And HOG and GIST are used as two types of features. 5) BBCSport5 is a collection of 544 documents associated with two views of sports articles from 5 subject areas. 6) Caltech-UCSD Birds (CUB)6 has 200 different categories, including 11788 images of birds with the corresponding textual descriptions (Reed et al. 2016). The image features are extracted by Goog Le Net, and the text features are extracted by Doc2Vec. The two kinds of features are used as different views. Footnotes provide URLs: 1https://archive.ics.uci.edu/ml/datasets/Multiple+Features 2https://www.cl.cam.ac.uk/research/dtg/attarchive/facedatabase 3http://www.cs.columbia.edu/CAVE/software/softlib/coil-20 4http://www.vision.caltech.edu/Image Datasets/Caltech101/ 5http://mlg.ucd.ie/datasets/ 6http://www.vision.caltech.edu/visipedia/CUB-200 |
| Dataset Splits | No | The paper describes datasets and mentions training, but does not provide specific details on how the training, validation, and test splits were performed (e.g., percentages, counts, or explicit splitting methodology). |
| Hardware Specification | No | The paper does not provide any specific hardware details such as CPU/GPU models, memory specifications, or cloud computing instance types used for running experiments. |
| Software Dependencies | No | The paper mentions using k-means algorithm but does not list any specific software libraries or dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x, TensorFlow 2.x). |
| Experiment Setup | No | The paper discusses model analysis related to dimensionality and convergence but does not provide specific experimental setup details such as hyperparameters (learning rate, batch size, number of epochs, optimizer settings) or training configurations for the models. |