MMD GAN: Towards Deeper Understanding of Moment Matching Network
Authors: Chun-Liang Li, Wei-Cheng Chang, Yu Cheng, Yiming Yang, Barnabas Poczos
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our evaluation on multiple benchmark datasets, including MNIST, CIFAR-10, Celeb A and LSUN, the performance of MMD GAN significantly outperforms GMMN, and is competitive with other representative GAN works. |
| Researcher Affiliation | Collaboration | Chun-Liang Li1, Wei-Cheng Chang1, Yu Cheng2 Yiming Yang1 Barnabás Póczos1 1 Carnegie Mellon University, 2AI Foundations, IBM Research {chunlial,wchang2,yiming,bapoczos}@cs.cmu.edu chengyu@us.ibm.com |
| Pseudocode | Yes | Algorithm 1: MMD GAN, our proposed algorithm. |
| Open Source Code | Yes | Our experiment code is available at https://github.com/October Chang/MMD-GAN. |
| Open Datasets | Yes | We train MMD GAN for image generation on the MNIST [32], CIFAR-10 [33], Celeb A [13], and LSUN bedrooms [12] datasets |
| Dataset Splits | Yes | We train MMD GAN for image generation on the MNIST [32], CIFAR-10 [33], Celeb A [13], and LSUN bedrooms [12] datasets. These are standard benchmark datasets which imply standard train/validation/test splits are used. |
| Hardware Specification | Yes | Figure 3 compares the computational time per generator iterations versus different B on Titan X. |
| Software Dependencies | No | The paper mentions using RMSProp [35] as an optimizer, but does not provide specific version numbers for any software libraries, frameworks, or programming languages used for implementation. |
| Experiment Setup | Yes | We use RMSProp [35] with learning rate of 0.00005 for a fair comparison with WGAN as suggested in its original paper [8]. We ensure the boundedness of model parameters of discriminator by clipping the weights point-wisely to the range [ 0.01, 0.01] as required by Assumption 2. The dimensionality h of the latent space is manually set according to the complexity of the dataset. We thus use h = 16 for MNIST, h = 64 for Celeb A, and h = 128 for CIFAR-10 and LSUN bedrooms. The batch size is set to be B = 64 for all datasets. |