Adversarial Ranking for Language Generation
Authors: Kevin Lin, Dianqi Li, Xiaodong He, Zhengyou Zhang, Ming-ting Sun
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on multiple public datasets clearly demonstrate the effectiveness of the proposed approach. |
| Researcher Affiliation | Collaboration | Kevin Lin University of Washington kvlin@uw.edu Dianqi Li University of Washington dianqili@uw.edu Xiaodong He Microsoft Research xiaohe@microsoft.com Zhengyou Zhang Microsoft Research zhang@microsoft.com Ming-Ting Sun University of Washington mts@uw.edu |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The synthetic data and the oracle model (LSTM model) are publicly available at https://github.com/Lantao Yu/Seq GAN. This refers to a third-party baseline, not the authors' own implementation code for Rank GAN. |
| Open Datasets | Yes | The synthetic data and the oracle model (LSTM model) are publicly available at https://github.com/Lantao Yu/Seq GAN. ... We conduct experiments on the Chinese poem dataset [37]... We test our method on the image captions provided by the COCO dataset [19]. ... We train our model on the Romeo and Juliet play [28]. |
| Dataset Splits | Yes | We randomly select 80, 000 captions as the training set, and select 5, 000 captions to form the validation set. |
| Hardware Specification | No | We thank NVIDIA Corporation for the donation of the GPU used for this research. This statement is too general and does not specify a particular GPU model or other detailed hardware specifications. |
| Software Dependencies | No | The paper does not list specific software components with their version numbers (e.g., Python 3.8, PyTorch 1.9). |
| Experiment Setup | Yes | we firstly collect 10, 000 sequential data generated by the oracle model (or true model) as the training set. ... Note that the sentence length of the training data is fixed to 20 for simplicity. ... After the standard pre-processing which replaces the non-frequently used words (appeared less than 5 times) with the special character UNK, we train our model on the dataset and generate the poem. ... We randomly select 80, 000 captions as the training set, and select 5, 000 captions to form the validation set. We replace the words appeared less than 5 times with UNK character. ... The script is splited into 2, 500 training sentences and 565 test sentences. To learn the rare words in the script, we adjust the threshold of UNK from 5 to 2. |