Research map / Probabilistic and scientific modeling / Molecules and proteins
Generative Inverse Design with Abstention via Diagonal Flow Matching
Miguel de Campos, Werner Krebs, Hanno Gottschalk
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 |
|---|---|---|---|
| anchor labels to zero and pair all design coordinates with noise (Diag-CFM instead of CFM) (ablation) | conditional flow matching · Gas Turbine Combustor | round-trip error: 1.85 × 10−2 -> 1.27 × 10−3 | Improved |
| zero-anchoring instead of standard CFM pairing (ablation) | conditional flow matching · Gas Turbine Combustor | forward MSE: 1.70 × 10−2 -> 1.39 × 10−3 | Improved |
| zero-anchoring instead of standard CFM pairing (ablation) | conditional flow matching · Unifoil | round-trip error: 1.10 × 10−2 -> 1.68 × 10−3 | Improved |
| zero-anchoring instead of standard CFM pairing (ablation) | conditional flow matching · DTLZ2 P=12 | round-trip error: 3.20 × 10−2 -> 9.10 × 10−4 | Improved |
| zero-anchoring instead of standard CFM pairing (ablation) | conditional flow matching · DTLZ2 P=100 | round-trip error: 4.52 × 10−2 -> 5.26 × 10−3 | Improved |
| zero-anchoring instead of standard CFM pairing (ablation) | conditional flow matching · FashionMNIST statistics (P=784) | forward MSE: 2.56 × 10−2 -> 7.20 × 10−4 | Improved |
| zero-anchoring instead of standard CFM pairing (ablation) | conditional flow matching · Thin film coating (P=128) | round-trip error: 2.902 × 10−2 -> 2.247 × 10−2 | Improved |
| permute order of design and label coordinates (ablation) | standard CFM vs Diag-CFM · Gas Turbine Combustor | final round-trip error | Hurt |
| use Diag-CFM instead of INN for forward prediction at high design dimension | Diag-CFM · DTLZ2 P=100 | forward MSE: 1.22 × 10−3 (INN) vs 9.58 × 10−3 (Diag-CFM) | Hurt |
| use Diag-CFM instead of CFM/INN for design diversity | Diag-CFM · Gas Turbine Combustor | design diversity: 5.27 × 10−2 (INN) vs 2.95 × 10−2 (Diag-CFM) | Hurt |
| select among K=10 generations with Zero-Deviation or Self-Consistency (ablation) | Diag-CFM · select-best | improvement over random selection: 5-31% over random | Improved |
| select among generations with Ensemble Variance or FM Loss (ablation) | Diag-CFM · select-best | improvement over random selection | Hurt |
Closest papers by meaning
- Compositional Generative Inverse Design (ICLR 2024 · similarity 0.87)
- 3DID: Direct 3D Inverse Design for Aerodynamics with Physics-Aware Optimization (NeurIPS 2025 · similarity 0.87)
- Solving Inverse Problems with FLAIR (NeurIPS 2025 · similarity 0.86)
- AFBench: A Large-scale Benchmark for Airfoil Design (NeurIPS 2024 · similarity 0.86)
- D-Flow: Differentiating through Flows for Controlled Generation (ICML 2024 · similarity 0.86)
- CcDPM: A Continuous Conditional Diffusion Probabilistic Model for Inverse Design (AAAI 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.
