Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
Authors: Alec Radford, Luke Metz, Soumith Chintala
ICLR 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks demonstrating their applicability as general image representations. |
| Researcher Affiliation | Industry | Alec Radford & Luke Metz indico Research Boston, MA {alec,luke}@indico.io Soumith Chintala Facebook AI Research New York, NY soumith@fb.com |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not include an unambiguous statement about releasing source code for the described methodology or a direct link to a code repository. |
| Open Datasets | Yes | We trained DCGANs on three datasets, Large-scale Scene Understanding (LSUN) (Yu et al., 2015), Imagenet-1k and a newly assembled Faces dataset. We use Imagenet-1k (Deng et al., 2009) as a source of natural images for unsupervised training. |
| Dataset Splits | Yes | On the Street View House Numbers dataset (SVHN)(Netzer et al., 2011), we split off a validation set of 10,000 examples from the non-extra set and use it for all hyperparameter and model selection. |
| Hardware Specification | Yes | Finally, we d like to thank Nvidia for donating a Titan-X GPU used in this work. |
| Software Dependencies | No | The paper mentions several software components like Adam optimizer, OpenCV, and various activation functions, but does not provide specific version numbers for any libraries or frameworks used in the implementation. |
| Experiment Setup | Yes | All models were trained with mini-batch stochastic gradient descent (SGD) with a mini-batch size of 128. All weights were initialized from a zero-centered Normal distribution with standard deviation 0.02. In the Leaky Re LU, the slope of the leak was set to 0.2 in all models. We used the Adam optimizer (Kingma & Ba, 2014) with tuned hyperparameters. We found the suggested learning rate of 0.001, to be too high, using 0.0002 instead. Additionally, we found leaving the momentum term β1 at the suggested value of 0.9 resulted in training oscillation and instability while reducing it to 0.5 helped stabilize training. |