Research map / Probabilistic and scientific modeling / Bayesian optimization
Improving black-box optimization in VAE latent space using decoder uncertainty
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 |
|---|---|---|---|
| Use decoder uncertainty censoring rather than unconstrained search | Character VAE gradient ascent · Arithmetic expressions | Validity: 60% ± 0.6% -> 98% ± 0.2% | Helped |
| Use decoder uncertainty censoring rather than unconstrained search | Character VAE gradient ascent · Arithmetic expressions | -log(1+MSE), Top 1: -0.32 ± 0.04 -> -0.13 ± 0.04 | Helped |
| Use decoder uncertainty censoring rather than unconstrained search | Character VAE Bayesian optimization · Arithmetic expressions | Validity: 77% ± 0.6% -> 98% ± 0.5% | Helped |
| Use decoder uncertainty censoring rather than unconstrained search | Character VAE Bayesian optimization · Arithmetic expressions | -log(1+MSE), Top 1: -0.57 ± 0.06 -> -0.41 ± 0.05 | Helped |
| Replace token-independence estimation with importance-sampling estimation (ablation) | Character VAE optimization · Molecule generation | decoder uncertainty identification of invalid decodings | Helped |
| Censor gradient-ascent proposals with IS-MI uncertainty | Character VAE · Molecules (ZINC) | Penalized logP, Top 1: 5.1 ± 0.2 -> 5.8 ± 0.2 | Helped |
| Use prior-probability censoring instead of decoder uncertainty censoring (ablation) | Character VAE gradient ascent · Molecules (ZINC) | Penalized logP, Top 1: 5.1 ± 0.2 -> 4.8 ± 0.1 | Hurt |
| Censor proposals with IS-MI uncertainty | Character VAE Bayesian optimization · Molecules (ZINC), search bounds 15 | Penalized logP, Top 1: 10.3 ± 3.9 -> 27.6 ± 2.1 | Helped |
| Censor proposals with IS-MI uncertainty | Character VAE Bayesian optimization · Molecules (ZINC), search bounds 10 | Validity: 1.1% ± 0.4% -> 10.6% ± 0.8% | Helped |
| Replace IS-MI with TI-MI uncertainty estimation (ablation) | Character VAE Bayesian optimization · Molecules (ZINC), search bounds 10 | Penalized logP, Top 1: IS-MI 6.6 ± 0.5; TI-MI 5.9 ± 3.3 | Mixed |
| Censor JT-VAE gradient-ascent proposals with IS-MI uncertainty | JT-VAE · Molecules (ZINC) | Quality top 10: 1% ± 1% -> 89% ± 3% | Helped |
| Censor JT-VAE proposals using prior probability rather than decoder uncertainty (ablation) | JT-VAE gradient ascent · Molecules (ZINC) | Penalized logP passing filters, Top 1: IS-MI 7.7 ± 0.7; NLLP 3.0 ± 0.1 | Hurt |
Closest papers by meaning
- Latent Bayesian Optimization via Autoregressive Normalizing Flows (ICLR 2025 · similarity 0.87)
- Return of the Latent Space COWBOYS: Re-thinking the use of VAEs for Bayesian Optimisation of Structured Spaces (ICML 2025 · similarity 0.87)
- Posterior Inference in Latent Space for Scalable Constrained Black-box Optimization (NeurIPS 2026 · similarity 0.87)
- PG-LBO: Enhancing High-Dimensional Bayesian Optimization with Pseudo-Label and Gaussian Process Guidance (AAAI 2024 · similarity 0.87)
- Be a Goldfish: Forgetting Bad Conditioning in Sparse Linear Regression via Variational Autoencoders (ICML 2025 · similarity 0.86)
- Inversion-based Latent Bayesian Optimization (NeurIPS 2024 · similarity 0.86)
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.