Research map / Image generation / GANs
MaskGAN: Better Text Generation via Filling in the _______
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train the generator adversarially on text infilling with an actor-critic objective instead of maximum likelihood alone | MaskMLE · IMDB reviews | overall human preference (%): 40.3 -> 44.3 | Helped |
| Train the generator adversarially on text infilling with an actor-critic objective instead of maximum likelihood alone | MaskMLE · PTB snippets | overall human preference (%): 31.3 -> 35.0 | Helped |
| Use MaskGAN rather than MaskMLE for unconditional generation | IMDB language model samples · IMDB samples under pretrained LM | perplexity: 273.1 $\pm$ 3.5 -> 108.3 $\pm$ 3.5 | Helped |
| Use MaskGAN instead of MaskMLE | IMDB review generation · IMDB reviews | unique bigrams (%): 43.6 -> 38.2 | Hurt |
| Use MaskGAN instead of MaskMLE | IMDB review generation · IMDB reviews | unique trigrams (%): 77.4 -> 70.7 | Hurt |
| Use MaskGAN instead of MaskMLE | IMDB review generation · IMDB reviews | unique quadgrams (%): 92.6 -> 88.2 | Hurt |
| Add a learned critic to the generator's policy-gradient training (ablation) | MaskGAN · Text generation training | gradient-estimate variance: an order of magnitude | Helped |
| Use attention in the in-filling generator (ablation) | MaskGAN · Text infilling | contextual plausibility | Helped |
| Use contiguous blocks of masked words rather than random masks (ablation) | MaskGAN · Text generation | sample quality | Helped |
| Give the discriminator the true masked context (ablation) | MaskGAN discriminator · Text infilling | accuracy of the generator learning signal | Helped |
Closest papers by meaning
- Branch-GAN: Improving Text Generation with (not so) Large Language Models (ICLR 2024 · similarity 0.92)
- Towards Text Generation with Adversarially Learned Neural Outlines (NeurIPS 2018 · similarity 0.91)
- Long Text Generation via Adversarial Training with Leaked Information (AAAI 2018 · similarity 0.89)
- Self-Adversarial Learning with Comparative Discrimination for Text Generation (ICLR 2020 · similarity 0.89)
- Language GANs Falling Short (ICLR 2020 · similarity 0.89)
- Adversarial Feature Matching for Text Generation (ICML 2017 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.