Unsupervised Cipher Cracking Using Discrete GANs
Authors: Aidan N. Gomez, Sicong Huang, Ivan Zhang, Bryan M. Li, Muhammad Osama, Lukasz Kaiser
ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate that Cipher GAN is capable of cracking language data enciphered using shift and Vigen ere ciphers to a high degree of fidelity and for vocabularies much larger than previously achieved. 4 EXPERIMENTS Our experiments use plaintext natural language samples from the Brown English text dataset (Francis & Kucera, 1979). |
| Researcher Affiliation | Collaboration | 1 Department of Computer Science, University of Toronto 2 FOR.ai 3 Google Brain |
| Pseudocode | No | The paper describes architectural details in Appendix A, but does not present them in pseudocode or a clearly labeled algorithm block. |
| Open Source Code | Yes | Code available at: github.com/for-ai/ciphergan |
| Open Datasets | Yes | Our experiments use plaintext natural language samples from the Brown English text dataset (Francis & Kucera, 1979). |
| Dataset Splits | No | The paper uses the Brown English text dataset and specifies batch size 64, but does not provide explicit train/validation/test dataset splits (e.g., percentages or sample counts) for reproducibility. |
| Hardware Specification | No | The paper does not provide specific details on the hardware used, such as exact GPU or CPU models, memory, or cloud instance types. |
| Software Dependencies | No | The paper mentions using the Adam optimizer and replacing batch normalization with layer normalization, but does not specify version numbers for any programming languages, libraries, or frameworks used for implementation. |
| Experiment Setup | Yes | We train using the Adam optmizer (Kingma & Ba, 2014) with batch size 64 and learning rate 2e 4, β1 = 0 and β2 = 0.9. Our learning rate is exponentially warmed up to 2e 4 over 2500 steps, and held constant thereafter. We use learned embedding vectors with 256 dimensions. The WGAN Lipschitz conditioning parameter was set to α = 10... |