Research map / Probabilistic and scientific modeling / Variational and generative sampling
On Variational Learning of Controllable Representations for Text without Supervision
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 |
|---|---|---|---|
| Constrain posterior means to a learned, filled probability simplex | CP-VAE · Yelp sentiment transfer | BLEU (BL): 27.91 ± 4.39 -> 33.69 ± 1.47 | Helped |
| Increase β-VAE manipulation magnitude from ±σ to the extremum | β-VAE · Yelp sentiment transfer | Accuracy (AC): 50.98 ± 2.89 -> 98.18 ± 1.56 | Helped |
| Increase β-VAE manipulation magnitude from ±σ to the extremum | β-VAE · Yelp sentiment transfer | BLEU (BL): 4.02 ± 0.77 -> 0.56 ± 0.40 | Hurt |
| Add the orthogonality regularizer L_REG (ablation) | CP-VAE · Yelp sentiment transfer | Accuracy (AC): 10.82 ± 0.91 -> 60.22 ± 4.57 | Helped |
| Add the structured reconstruction loss L_S-REC (ablation) | CP-VAE · Yelp sentiment transfer | Accuracy (AC): 12.28 ± 3.69 -> 60.22 ± 4.57 | Helped |
| Add the structured reconstruction loss L_S-REC (ablation) | CP-VAE · Yelp sentiment transfer | BLEU (BL): 49.34 ± 2.65 -> 33.69 ± 1.47 | Hurt |
| Use CP-VAE instead of adversarial baselines or D&R | Text style transfer · Yelp | GLEU (GL): CP-G: 7.5; D&R: 7.9 | Mixed |
| Use CP-VAE instead of LDA | AG News topic identification · World | F1: 72.14 -> 74.59 | Helped |
| Use CP-VAE instead of LDA | AG News topic identification · Sports | F1: 80.22 -> 79.49 | Hurt |
| Use CP-VAE instead of LDA | AG News topic identification · Business | F1: 68.46 -> 63.97 | Hurt |
| Use CP-VAE instead of LDA | AG News topic identification · Sci/Tech | F1: 61.60 -> 66.77 | Helped |
Closest papers by meaning
- Improved Variational Autoencoders for Text Modeling using Dilated Convolutions (ICML 2017 · similarity 0.89)
- Adversarially Regularized Autoencoders (ICML 2018 · similarity 0.88)
- Educating Text Autoencoders: Latent Representation Guidance via Denoising (ICML 2020 · similarity 0.88)
- Constraining Variational Inference with Geometric Jensen-Shannon Divergence (NeurIPS 2020 · similarity 0.87)
- Improving Variational Autoencoders with Density Gap-based Regularization (NeurIPS 2022 · similarity 0.87)
- Learning Latent Subspaces in Variational Autoencoders (NeurIPS 2018 · similarity 0.87)
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.