Non-Parametric Priors For Generative Adversarial Networks
Authors: Rajhans Singh, Pavan Turaga, Suren Jayasuriya, Ravi Garg, Martin Braun
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To validate the effectiveness of the proposed approach, we train the standard DCGAN model (Radford et al., 2016) on four different datasets: a) Celeb A dataset (Liu et al., 2015), b) CIFAR10 (Krizhevsky & Hinton, 2009), c) LSUN Bedroom, and d) LSUN Kitchen (Yu et al., 2015)). We train our model to the same number of epochs and all the hyper-parameters of the training are kept same for all the cases. We train each model three times and report the average scores. Details about the network architecture and the training method are provided in the supplemental material. We compare our proposed non-parametric prior distribution against standard ones like the normal, uniform and the priors designed to minimize the mid-point distribution mismatch like Gamma (Kilcher et al., 2018), and Cauchy (Le sniak et al., 2019). For Gamma and Cauchy, we use the same parameters as suggested in the corresponding references. Qualitative tests: In Figure 5, we show the effect of interpolation through origin in high latent-space dimension (d = 100) for different priors on Celeb A datatset. Quantitative evaluation: For quantitative analysis, we use the Inception Score (IS) (Salimans et al., 2016) and the Frechet Inception Distance (FID) (Heusel et al., 2017) which are the standard metrics used to evaluate GAN performance. We will see in Tables 2 to 6, that our non-parametric prior performs better in terms of FID score on both the prior and mid-point by at least 2 points. |
| Researcher Affiliation | Collaboration | 1School of Electrical, Computer, and Energy Engineering, Arizona State University, Tempe, AZ, USA 2School of Arts, Media and Engineering, Arizona State University, Tempe, AZ, USA 3Intel Corporation, Chandler, AZ, USA. Correspondence to: Rajhans Singh <rsingh70@asu.edu>, Pavan Turaga <pturaga@asu.edu>, Suren Jayasuriya <sjayasur@asu.edu>, Ravi Garg <ravi.garg@intel.com>, Martin W. Braun <martin.w.braun@intel.com>. |
| Pseudocode | No | The paper describes the optimization problem and how it's solved using fmincon in Matlab, but it does not provide any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code for their method or links to a repository. |
| Open Datasets | Yes | a) Celeb A dataset (Liu et al., 2015), b) CIFAR10 (Krizhevsky & Hinton, 2009), c) LSUN Bedroom, and d) LSUN Kitchen (Yu et al., 2015)). |
| Dataset Splits | No | The paper states 'We train our model to the same number of epochs and all the hyper-parameters of the training are kept same for all the cases.' and mentions 'Details about the network architecture and the training method are provided in the supplemental material.' It does not explicitly specify train/validation/test splits within the main text. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware used to run its experiments, such as GPU models, CPU types, or memory details. |
| Software Dependencies | No | The paper mentions 'fmincon in Matlab' but does not specify the version number for Matlab or any other software dependencies with version numbers. |
| Experiment Setup | Yes | We solve (7) using fmincon in Matlab, which uses an interior-point algorithm with barrier functions. We note that the solution from fmincon may only be a locally optimal solution, yet we find the obtained solution is quite robust to large variation in initialization. We also note that using any of KL(P Q), KL(Q P) and KL(P (P +Q) 2 ) + KL(Q (P +Q) 2 ) as objective in (7) gives us the same result. We use the following settings for fmincon: interior-point as algorithm, Max Function Evaluations = 4 105, Max Iterations = 105 and n = 1024. We use ξ = 0.75 in our experiments because it provides the best FID score (Heusel et al., 2017). |