Variational Laplace Autoencoders
Authors: Yookoon Park, Chris Kim, Gunhee Kim
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results on MNIST, Omniglot, Fashion-MNIST, SVHN and CIFAR10 show that the proposed approach significantly outperforms other recent amortized or iterative methods on the Re LU networks. (From Abstract) ... We evaluate our approach on five popular datasets: MNIST, Omniglot, Fashion-MNIST, SVHN and CIFAR-10. (From Section 5. Experiments) |
| Researcher Affiliation | Academia | Yookoon Park 1 Chris Dongjoo Kim 1 Gunhee Kim 1 1Neural Processing Research Center, Seoul National University, Seoul, South Korea. |
| Pseudocode | Yes | Algorithm 1 Posterior inference for piece-wise linear nets; Algorithm 2 Variational Laplace Autoencoders |
| Open Source Code | Yes | The code is public at http://vision.snu.ac.kr/projects/VLAE. |
| Open Datasets | Yes | We evaluate our approach on five popular datasets: MNIST (Le Cun et al., 1998), Omniglot (Lake et al., 2013), Fashion MNIST (Xiao et al., 2017), Street View House Numbers (SVHN) (Wang et al., 2011) and CIFAR-10 (Krizhevsky & Hinton, 2009). |
| Dataset Splits | No | The paper refers to training, validation, and test sets conceptually, but does not provide specific details on the dataset splits (e.g., percentages or sample counts) used for the experiments. |
| Hardware Specification | Yes | For example, our MNIST model takes 10 GPU hours for 1,000 epochs on one Titan X Pascal. |
| Software Dependencies | No | The paper discusses algorithms and optimizers but does not specify any software names with version numbers (e.g., "PyTorch 1.9" or "Python 3.8"). |
| Experiment Setup | Yes | We experiment two network settings: (1) a small network with one hidden layer. The latent variable dimension is 16 and the hidden layer dimension is 256. We double both dimensions for color datasets of SVHN and CIFAR10. (2) A bigger network with two hidden layers. The latent variable dimension is 50 and the hidden layer dimension is 500 for all datasets. For both settings, we apply Re LU activation to hidden layers and use the same architecture for the encoder and decoder. |