Research map / Architectures and efficiency / RNNs and normalization
Learning Robust Hierarchical Patterns of Human Brain across Many fMRI Studies
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| jointly model site-specific components and adversarially reduce site information | hSCP on simulated one-level fMRI · Simulated one-level, k1=10 | component accuracy: 0.787 -> 0.910 | Helped |
| jointly model site effects | hSCP on simulated one-level fMRI · Simulated one-level, k1=10 | component accuracy: 0.787 -> 0.865 | Helped |
| add adversarial learning to rshSCP (ablation) | rshSCP on simulated one-level fMRI · Simulated one-level, k1=10 | component accuracy: 0.865 -> 0.910 | Helped |
| replace joint estimation with two-step ComBat harmonization | hSCP on simulated one-level fMRI · Simulated one-level, k1=10 | component accuracy: 0.787 -> 0.759 | Hurt |
| add adversarial learning to rshSCP (ablation) | rshSCP on real multi-site fMRI · Real multi-site fMRI, k2=4, k1=10 | split-sample reproducibility: 0.806±0.036 -> 0.808±0.030 | Helped |
| add joint site-effect modeling | Adv. hSCP on real multi-site fMRI · Real multi-site fMRI, k2=4, k1=10 | leave-one-site-out reproducibility: 0.656±0.035 -> 0.716±0.032 | Helped |
| replace SVD-based initialization with random initialization (ablation) | rshSCP on simulated one-level fMRI · Simulated one-level, k1=10 | component accuracy: 0.865 -> 0.834±0.055 | Hurt |
| replace SVD-based initialization with random initialization (ablation) | Adv. rshSCP on simulated one-level fMRI · Simulated one-level, k1=10 | component accuracy: 0.910 -> 0.895±0.036 | Hurt |
| reduce site predictability from subject-specific features | Adv. rshSCP on simulated data · Simulated one-level | site-prediction accuracy (SVM): 97.6% -> 67% | Hurt |
| reduce site predictability from subject-specific features | rshSCP on real multi-site fMRI · Real multi-site fMRI | site-prediction accuracy (SVM): 51% to 32% | Hurt |
| apply rshSCP with adversarial learning instead of hSCP | age prediction from subject-specific features · Real multi-site fMRI, k2=4, k1=10 | age prediction MAE: 6.490± 1.485 -> 6.494± 1.501 | No clear effect |
| vary site-factor sparsity parameter μ (ablation) | rshSCP on real dataset · Real dataset, one-level, k1=10 | component accuracy: 0.756; 0.865; 0.789 | Mixed |
Closest papers by meaning
- Parameter-Free Centralized Multi-Task Learning for Characterizing Developmental Sex Differences in Resting State Functional Connectivity (AAAI 2018 · similarity 0.85)
- Testing for Differences in Gaussian Graphical Models: Applications to Brain Connectivity (NeurIPS 2016 · similarity 0.85)
- Large Connectome Model: An fMRI Foundation Model of Brain Connectomes Empowered by Brain-Environment Interaction in Multitask Learning Landscape (AAAI 2026 · similarity 0.85)
- Learning brain regions via large-scale online structured sparse dictionary learning (NeurIPS 2016 · similarity 0.84)
- Learning High-Order Relationships of Brain Regions (ICML 2024 · similarity 0.84)
- FAST-Brain: A Flow-Aligned Spatio-Temporal Surrogate Brain Model (NeurIPS 2026 · 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.