Research map / Learning theory and algorithms / Kernels and regression theory
Certificates for Complex-Compatible Learned Cochain Laplacians
Nivar Anwer, Marien Chenaud, David Elizondo
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 soft defect penalty (ablation) | learned cochain Laplacian · Poisson PDE1 (N=32) | condition number of L0: (1.73 ± 0.29) × 104 -> (2.47 ± 0.12) × 102 | Improved |
| add soft defect penalty (ablation) | learned cochain Laplacian · Poisson PDE1 (N=64) | condition number of L0: (3.58 ± 1.47) × 106 -> (9.63 ± 0.28) × 102 | Improved |
| add soft defect penalty (ablation) | learned cochain Laplacian · Poisson PDE1 (N=64) | test MSE: (1.94 ± 0.81) × 104 -> (4.83 ± 0.27) × 10−8 | Improved |
| replace soft penalty with hard chain-compatible parameterisation (ablation) | learned cochain Laplacian · Poisson PDE1 (N=64) | test MSE: (4.83 ± 0.27) × 10−8 -> (7.21 ± 0.48) × 10−10 | Improved |
| replace soft penalty with hard chain-compatible parameterisation (ablation) | learned cochain Laplacian · Poisson PDE1 (N=64) | condition number of L0: (9.63 ± 0.28) × 102 -> (9.58 ± 0.19) × 102 | No clear effect |
| add soft defect regularisation (ablation) | learned cochain Laplacian · Topology shift (beta0 in {1,2,3}) | Betti accuracy: 0.71 -> 0.93 | Improved |
| use hard chain-compatible parameterisation (ablation) | learned cochain Laplacian · Topology shift (beta0 in {1,2,3}) | Betti accuracy: 0.71 -> 0.98 | Improved |
| use hard chain-compatible parameterisation (ablation) | learned cochain Laplacian · Topology shift (beta0 in {1,2,3}) | nRMSE: 4.6×10−2 -> 4.3×10−2 | No clear effect |
| raise soft defect weight lambda from 0.1 to 10 (ablation) | learned cochain Laplacian · Topology-shift OOD (beta0 in {2,3,4}) | Betti error rate: 0.094 ± 0.024 -> 0.021 ± 0.011 | Improved |
| use manifold (hard) constraint instead of defect control (ablation) | learned cochain Laplacian · Topology-shift OOD (beta0 in {2,3,4}) | mean OOD PDE error (nRMSE): 0.731 ± 0.075 -> 0.832 ± 0.094 | Hurt |
| add post-hoc projection to unconstrained model (ablation) | learned cochain Laplacian · Topology-shift OOD (beta0 in {2,3,4}) | mean OOD PDE error (nRMSE): 0.763 ± 0.082 -> 1.186 ± 0.217 | Hurt |
| add full defect-controlled training (ablation) | learned cochain Laplacian · Topology-shift OOD (beta0 in {2,3,4}) | Betti error rate: 0.147 ± 0.031 -> 0.0047 ± 0.0041 | Improved |
Closest papers by meaning
- MC^2: Monte Carlo Correction for Fast Elliptic PDE Solving (NeurIPS 2026 · similarity 0.83)
- Neural Spectral Methods: Self-supervised learning in the spectral domain (ICLR 2024 · similarity 0.83)
- Holistic Physics Solver: Learning PDEs in a Unified Spectral-Physical Space (ICML 2025 · similarity 0.83)
- Learning to Discover Iterative Spectral Algorithms (NeurIPS 2026 · similarity 0.82)
- Spectral Preconditioning for Gradient Methods on Graded Non-convex Functions (ICML 2024 · similarity 0.82)
- A Neural-Preconditioned Poisson Solver for Mixed Dirichlet and Neumann Boundary Conditions (ICML 2024 · similarity 0.82)
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.
