Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Structured SIR: Efficient and Expressive Importance-Weighted Inference for High-Dimensional Image Registration
Ivor Simpson, Neill Campbell
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 SIR training instead of variational (ablation) | LC covariance · OASIS label calibration | AUSE label: 0.0844 -> 0.0659 | Improved |
| add SIR training instead of variational (ablation) | LC covariance · OASIS label calibration | ECE: 0.1274 -> 0.1064 | Improved |
| use naive importance sampling instead of variational (ablation) | LC covariance · OASIS label calibration | AUSE label: 0.0844 -> 0.0861 | No clear effect |
| add low-rank component to SIR with sparse Cholesky (ablation) | S+C · OASIS label calibration | AUSE label: 0.0907 -> 0.0659 | Improved |
| add sparse Cholesky to SIR with low-rank (ablation) | S+LD · OASIS label calibration | AUSE label: 0.0736 -> 0.0659 | Improved |
| add structured covariance in variational inference (ablation) | V+D · OASIS label calibration | AUSE label: 0.0850 -> 0.0844 | No clear effect |
| add structured covariance in SIR (ablation) | S+D · OASIS registration | Oracle DSC: 0.805 -> 0.812 | Improved |
| use SIR instead of variational (mean prediction) (ablation) | LC covariance · OASIS registration | fold % of mean: 0.044 -> 0.011 | Improved |
| use SIR instead of variational (resampled fields) (ablation) | LC covariance · OASIS registration | fold % of resampled fields: 0.056 -> 0.115 | Hurt |
| reduce low-rank components from 25 to 10 (ablation) | S+LC · OASIS label calibration | AUSE label: 0.0659 -> 0.0636 | Improved |
| increase low-rank components from 25 to 30 (ablation) | S+LC · OASIS label calibration | AUSE label: 0.0659 -> 0.0725 | Hurt |
| raise temperature from 2.0 to 3.0 (ablation) | S+LC · OASIS label calibration | AUSE label: 0.0775 -> 0.0659 | Improved |
Closest papers by meaning
- Uncertainty Estimation for Pretrained Medical Image Registration Models via Transformation Equivariance (NeurIPS 2026 · similarity 0.87)
- Hierarchical Uncertainty Estimation for Learning-based Registration in Neuroimaging (ICLR 2025 · similarity 0.86)
- CARL: A Framework for Equivariant Image Registration (CVPR 2025 · similarity 0.85)
- Scalable 3D Registration via Truncated Entry-wise Absolute Residuals (CVPR 2024 · similarity 0.85)
- Variational Multi-scale Representation for Estimating Uncertainty in 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.85)
- Sparse-to-dense Multimodal Image Registration via Multi-Task Learning (ICML 2024 · similarity 0.84)
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.
