Research map / Probabilistic and scientific modeling / Variational and generative sampling
Grammar Variational Autoencoder
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 |
|---|---|---|---|
| Represent sequences as grammar production rules and decode with grammar-valid rule masks instead of decoding characters | VAE for arithmetic expressions · Arithmetic expressions (Bayesian optimization) | fraction valid: 0.86 ± 0.06 -> 0.99 ± 0.01 | Helped |
| Represent sequences as grammar production rules and decode with grammar-valid rule masks instead of decoding characters | VAE for arithmetic expressions · Arithmetic expressions (Bayesian optimization) | average score (log(1 + MSE)): 4.75 ± 0.25 -> 3.47 ± 0.24 | Helped |
| Represent sequences as grammar production rules and decode with grammar-valid rule masks instead of decoding characters | VAE for molecules · Molecules (Bayesian optimization) | fraction valid: 0.17 ± 0.05 -> 0.31 ± 0.07 | Helped |
| Represent sequences as grammar production rules and decode with grammar-valid rule masks instead of decoding characters | VAE for molecules · Molecules (Bayesian optimization) | average penalized logP score: -54.66 ± 2.66 -> -9.57 ± 1.77 | Helped |
| Use grammar-tree latent representations rather than character-sequence representations | Sparse GP property predictor · Expressions (latent features; test split) | test log-likelihood: -1.397 ± 0.003 -> -1.320 ± 0.001 | Helped |
| Use grammar-tree latent representations rather than character-sequence representations | Sparse GP property predictor · Molecules (latent features; test split) | test log-likelihood: -1.812 ± 0.004 -> -1.739 ± 0.004 | Helped |
| Use grammar-tree latent representations rather than character-sequence representations | Sparse GP property predictor · Expressions (latent features; test split) | RMSE: 0.975 ± 0.004 -> 0.884 ± 0.002 | Helped |
| Use grammar-tree latent representations rather than character-sequence representations | Sparse GP property predictor · Molecules (latent features; test split) | RMSE: 1.504 ± 0.006 -> 1.404 ± 0.006 | Helped |
| Encode and decode molecules using grammar production rules rather than characters | Molecule VAE · Molecules (hold-out reconstruction) | reconstruction accuracy (%): 44.6 -> 53.7 | Helped |
| Encode and decode molecules using grammar production rules rather than characters | Molecule VAE · Molecules (prior samples) | prior validity (%): 0.70 -> 7.2 | Helped |
Closest papers by meaning
- Syntax-Directed Variational Autoencoder for Structured Data (ICLR 2018 · similarity 0.92)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.89)
- NeVAE: A Deep Generative Model for Molecular Graphs (AAAI 2019 · similarity 0.88)
- Constrained Graph Variational Autoencoders for Molecule Design (NeurIPS 2018 · similarity 0.88)
- Discrete Object Generation with Reversible Inductive Construction (NeurIPS 2019 · similarity 0.88)
- Data-Efficient Graph Grammar Learning for Molecular Generation (ICLR 2022 · 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.