Research map / Probabilistic and scientific modeling / Bayesian optimization
Sample-Efficient Optimization in the Latent Space of Deep Generative Models via Weighted Retraining
Austin Tripp, Erik Daxberger, José Miguel Hernández-Lobato
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Add weighted retraining (rank weighting, k=10^-3, with Bayesian optimization) | JT-VAE · Chemical design (ZINC250k), 500 queries | top 1 penalized logP score: 5.65 -> 27.84 | Improved |
| Add retraining to rank weighting (ablation) | JT-VAE · Chemical design (ZINC250k), 500 queries | top 1 penalized logP score: 5.95 -> 21.20 | Improved |
| Use rank weighting without retraining (ablation) | JT-VAE · Chemical design (ZINC250k), 500 queries | top 1 penalized logP score: 5.65 -> 5.95 | Improved |
| Combine rank weighting and periodic retraining rather than using either alone (ablation) | Latent-space optimization · 2D shapes, arithmetic expressions, and chemical design | TopK score | Improved |
| Increase weighting and retraining frequency (lower k and r) (ablation) | Latent-space optimization · 2D shapes and chemical design | TopK score | Improved |
| Increase retraining frequency (lower r) (ablation) | Latent-space optimization · Arithmetic expression fitting | TopK score | Mixed |
| Use weighted retraining instead of cross-entropy-based baselines | Latent-space optimization · 2D shapes, arithmetic expressions, and chemical design | TopK score | Improved |
| Use weighted retraining instead of MolDQN | Chemical design · ZINC250k, 500 versus at least 5000 samples | top 1 penalized logP score: 11.84 -> 27.84 | Improved |
| Use weighted retraining instead of ChemBO at 100 samples | Chemical design · ZINC250k, 100 queries | penalized logP score | Hurt |
| Train the generative model with rank-based weighting rather than uniform weighting (ablation) | Deep generative models · 2D shapes, arithmetic expressions, and chemical design | objective value of prior samples | Improved |
Closest papers by meaning
- Posterior Inference in Latent Space for Scalable Constrained Black-box Optimization (NeurIPS 2026 · similarity 0.88)
- Self-Consuming Generative Models with Curated Data Provably Optimize Human Preferences (NeurIPS 2024 · similarity 0.87)
- Generative Adversarial Model-Based Optimization via Source Critic Regularization (NeurIPS 2024 · similarity 0.86)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.85)
- A Diffusion Model Framework for Unsupervised Neural Combinatorial Optimization (ICML 2024 · similarity 0.85)
- On the Stability of Iterative Retraining of Generative Models on their own Data (ICLR 2024 · similarity 0.85)
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.
