Research map / Probabilistic and scientific modeling / Bayesian optimization
Bidirectional Learning for Offline Infinite-width Model-based Optimization
Can Chen, Yingxue Zhang, Jie Fu, Xue Liu, Mark Coates
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 |
|---|---|---|---|
| Use bidirectional learning with the NTK model rather than the strongest reported baseline | Offline model-based optimization · Superconductor | 100th percentile normalized ground truth score: 0.488 ± 0.034 -> 0.520 ± 0.005 | Improved |
| Use bidirectional learning with the NTK model rather than the strongest reported baseline | Offline model-based optimization · Ant Morphology | 100th percentile normalized ground truth score: 1.214 ± 0.732 -> 0.962 ± 0.000 | Hurt |
| Use bidirectional learning with the NTK model rather than the strongest reported baseline | Offline model-based optimization · D’Kitty Morphology | 100th percentile normalized ground truth score: 0.942 ± 0.010 -> 0.941 ± 0.000 | No clear effect |
| Use bidirectional learning with the NTK model rather than the strongest reported baseline | Offline model-based optimization · Hopper Controller | 100th percentile normalized ground truth score: 1.959 ± 0.159 -> 1.989 ± 0.050 | Improved |
| Use bidirectional learning with the NTK model rather than the strongest reported baseline | Offline model-based optimization · GFP | 100th percentile normalized ground truth score: 0.865 ± 0.000 -> 0.864 ± 0.000 | No clear effect |
| Use bidirectional learning with the NTK model rather than the strongest reported baseline | Offline model-based optimization · TF Bind 8 | 100th percentile normalized ground truth score: 0.953 ± 0.022 -> 0.973 ± 0.000 | Improved |
| Use bidirectional learning with the NTK model rather than the strongest reported baseline | Offline model-based optimization · UTR | 100th percentile normalized ground truth score: 0.699 ± 0.011 -> 0.760 ± 0.000 | Improved |
| Add the forward mapping loss (ablation) | BDI · TF Bind 8 | 100th percentile normalized ground truth score: 0.984 ± 0.000 -> 0.973 ± 0.000 | Hurt |
| Add the forward mapping loss (ablation) | BDI · GFP, UTR, Superconductor, Ant Morphology, D’Kitty Morphology, Hopper Controller | 100th percentile normalized ground truth score: GFP: 0.860 ± 0.004 -> 0.864 ± 0.000; UTR: 0.636 ± 0.000 -> 0.760 ± 0.000; SuperC: 0.511 ± 0.007 -> 0.520 ± 0.005; Ant: 0 | Improved |
| Add the backward mapping loss (ablation) | BDI · UTR | 100th percentile normalized ground truth score: 0.777 ± 0.000 -> 0.760 ± 0.000 | Hurt |
| Add the backward mapping loss (ablation) | BDI · Superconductor, Ant Morphology, Hopper Controller | 100th percentile normalized ground truth score: SuperC: 0.502 ± 0.007 -> 0.520 ± 0.005; Ant: 0.933 ± 0.000 -> 0.962 ± 0.000; Hopper: 1.821 ± 0.223 -> 1.989 ± 0.050 | Improved |
| Add the backward mapping loss (ablation) | BDI · GFP, TF Bind 8, D’Kitty Morphology | 100th percentile normalized ground truth score: GFP: 0.864 ± 0.000 -> 0.864 ± 0.000; TFB: 0.973 ± 0.000 -> 0.973 ± 0.000; D’Kitty: 0.942 ± 0.001 -> 0.941 ± 0.000 | No clear effect |
Closest papers by meaning
- Offline Model-Based Optimization via Policy-Guided Gradient Search (AAAI 2024 · similarity 0.88)
- Learning Surrogates for Offline Black-Box Optimization via Gradient Matching (ICML 2024 · similarity 0.88)
- Towards Universal Offline Black-Box Optimization via Learning Language Model Embeddings (ICML 2025 · similarity 0.86)
- Offline Model-Based Optimization by Learning to Rank (ICLR 2025 · similarity 0.86)
- Generative Adversarial Model-Based Optimization via Source Critic Regularization (NeurIPS 2024 · similarity 0.86)
- Incorporating Surrogate Gradient Norm to Improve Offline Optimization Techniques (NeurIPS 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.
