Toward Controlled Generation of Text

Authors: Zhiting Hu, Zichao Yang, Xiaodan Liang, Ruslan Salakhutdinov, Eric P. Xing

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Quantitative experiments using trained classifiers as evaluators validate the accuracy of sentence and attribute generation.
Researcher Affiliation Collaboration 1Carnegie Mellon University 2Petuum, Inc.. Correspondence to: Zhiting Hu <zhitingh@cs.cmu.edu>.
Pseudocode Yes Algorithm 1 Controlled Generation of Text
Open Source Code No The paper does not provide any explicit statement or link regarding the release of open-source code for the described methodology.
Open Datasets Yes We use a large IMDB text corpus (Diao et al., 2014) for training the generative models. Stanford Sentiment Treebank-2 (SST-full) (Socher et al., 2013) consists of 6920/872/1821 movie review sentences with binary sentiment annotations in the train/dev/test sets, respectively. Lexicon from (Wilson et al., 2005). We compile from the Time Bank (timeml.org) dataset.
Dataset Splits Yes Stanford Sentiment Treebank-2 (SST-full) (Socher et al., 2013) consists of 6920/872/1821 movie review sentences with binary sentiment annotations in the train/dev/test sets, respectively.
Hardware Specification No The paper does not explicitly describe the specific hardware used (e.g., GPU/CPU models, memory) for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., library names like PyTorch or TensorFlow with their versions).
Experiment Setup Yes The generator and encoder are set as single-layer LSTM RNNs with input/hidden dimension of 300 and max sample length of 15. Discriminators are set as Conv Nets. ... we use a KL term weight linearly annealing from 0 to 1 during training. Balancing parameters are set to λc = λz = λu = 0.1, and β is selected on the dev sets.