Tensor-Variate Restricted Boltzmann Machines
Authors: Tu Nguyen, Truyen Tran, Dinh Phung, Svetha Venkatesh
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the capacity of Tv RBMs on three real-world applications: handwritten digit classification, face recognition and EEG-based alcoholic diagnosis. The learnt features of the model are more discriminative than the rivals, resulting in better classification performance. The experimental results show that the classification performance of the Tv RBM is more competitive than the standard RBM and existing tensor decomposition methods. |
| Researcher Affiliation | Academia | Tu Dinh Nguyen , Truyen Tran , Dinh Phung , Svetha Venkatesh Center for Pattern Recognition and Data Analytics School of Information Technology, Deakin University, Geelong, Australia Institute for Multi-Sensor Processing and Content Analysis Curtin University, Australia |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | We use the MNIST dataset which consists of 60, 000 training and 10, 000 test images of digits from 0 to 9 with the size 28 28. We use Extended Yale B database (Lee, Ho, and Kriegman 2005) which contains images of 28 subjects under 65 illumination conditions (including the ambient lighting) and 9 poses. The EEG dataset collected in (Zhang et al. 1995) contains readings of 64 electrodes placed on the scalp of 122 subjects in two (alcoholic and control) groups. |
| Dataset Splits | No | We use the MNIST dataset which consists of 60, 000 training and 10, 000 test images of digits from 0 to 9 with the size 28 28. Hyperparameters are specified using cross-validation. |
| Hardware Specification | No | The paper does not provide specific hardware details (like CPU/GPU models, memory, or cloud instances) used for running the experiments. |
| Software Dependencies | No | The paper mentions 'The n-way toolbox for matlab' for Tucker and PARAFAC implementations, but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | The pixel intensities of images are normalized into the range [0, 1] and treated as empirical probabilities of the input. The factorized parameters are initialized randomly from N (0; 0.1). For binary models, the step size of is fixed to 0.01. Hidden, visible and mode-factor learning rates are fixed to 0.1. In the Gaussian models, the step size and learning rates are 10 times smaller due to unbounded visible variables. The number of factors F of Tv RBM is set to 100. For both RBM and Tv RBM, 500 hidden units are used for images and 200 for EEG signals. Hyperparameters are specified using cross-validation. We update parameters after seeing mini-batches of B = 50 samples. Learning is terminated after 100 scans through the whole data. |