Calibrating Energy-based Generative Adversarial Networks
Authors: Zihang Dai, Amjad Almahairi, Philip Bachman, Eduard Hovy, Aaron Courville
ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, the experiment results closely match our theoretical analysis, verifying that the discriminator is able to recover the energy of data distribution. In this section, we verify our theoretical results empirically on several synthetic and real datasets. [...] Experiment results are summarized in Figure 2 for baseline models, and Figure 3 for the proposed models. |
| Researcher Affiliation | Collaboration | Zihang Dai1, Amjad Almahairi2 , Philip Bachman3, Eduard Hovy1 & Aaron Courville2 1 Language Technologies Institute, Carnegie Mellon University. 2 MILA, Universit e de Montr eal. 3 Maluuba Research. |
| Pseudocode | No | The paper contains mathematical derivations and equations but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | For more details, please refer to https://github.com/zihangdai/cegan_iclr2017. |
| Open Datasets | Yes | We train on 28 28 images of a single handwritten digit from the NIST dataset. https://www.nist.gov/srd/nist-special-database-19, which is an extended version of MNIST with an average of over 74K examples per digit. [...] In this last set of experiments, we evaluate the visual quality of samples generated by our model in two datasets of natural images, namely CIFAR-10 and Celeb A. |
| Dataset Splits | No | The paper mentions '100K training samples' for synthetic data and 'real test images' for NIST digits, but does not specify explicit training/validation/test dataset splits (e.g., percentages, sample counts for each split, or detailed splitting methodology). |
| Hardware Specification | No | The paper provides details on neural network architectures and optimization algorithms but does not specify any hardware components (e.g., GPU or CPU models) used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'Theano (Theano Development Team, 2016)' and 'Adam as the optimization algorithm', but it does not provide specific version numbers for these or any other software dependencies like Python, PyTorch, or TensorFlow versions. |
| Experiment Setup | Yes | For experiments with the synthetic datasets, the following fully-connected feed forward neural networks are employed Generator: FC(4,128)-BN-Re LU-FC(128,128)-BN-Re LU-FC(128,2) Discriminator: FC(2,128)-Re LU-FC(128,128)-Re LU-FC(128,1) Inference Net: FC(2,128)-Re LU-FC(128,128)-Re LU-FC(128,4*2). [...] Given the chosen architectures, we follow Radford et al. (2015) and use Adam as the optimization algorithm. |