Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Neural Representations of Human Cognition across Many fMRI Studies
Arthur Mensch, Julien Mairal, Danilo Bzdok, Bertrand Thirion, Gael Varoquaux
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 |
|---|---|---|---|
| Project brain maps onto resting-state functional-network dictionaries (ablation) | Multinomial classification · Four target datasets | test accuracy | Improved |
| Use input-space Dropout instead of ℓ2 regularization (ablation) | Dictionary-projected multinomial classifier · Four target datasets | test accuracy: +1% to +5% | Improved |
| Learn a second, cognitive-aware projection and apply Dropout in that latent space (ablation) | Dictionary-projected classifier · Three of four target datasets | test accuracy: +3% to +7% | Improved |
| Train the shared cognitive projection using multiple datasets instead of only the target study (ablation) | Factored model · Three of four target studies | test accuracy: +1.1% to +1.6% | Improved |
| Train the shared cognitive projection using other studies (ablation) | Factored model · LA5c | test accuracy | Hurt |
| Use multi-scale dictionaries rather than a single 512-component dictionary (ablation) | Multi-dataset cognitive-space model · Four target datasets | predictive accuracy | Improved |
| Use multi-scale dictionaries rather than a single 128-component dictionary (ablation) | Multi-dataset cognitive-space model · Four target datasets | predictive accuracy | Improved |
| Use a latent cognitive space with Dropout rather than omit the latent projection (ablation) | Factored model · Archi, Brainomics, CamCan | test accuracy: +3% to +7% | Improved |
| Increase the target training set size by adding training subjects (ablation) | Single-study and multi-dataset models · Archi, Brainomics, CamCan | test accuracy | Mixed |
| Use a multi-scale dictionary rather than a single-scale dictionary (ablation) | Multi-dataset cognitive-space model · Four target datasets, 512-component dictionary comparison | predictive accuracy | No clear effect |
| Replace explicit factorization with trace-norm-regularized convex multi-task learning (ablation) | Multi-dataset classification model · Task fMRI datasets | test accuracy | No clear effect |
| Increase the dropout rate from 0.5 to 0.75 (ablation) | Latent cognitive-space model · Task fMRI datasets | test accuracy | Improved |
Closest papers by meaning
- Neural Taskonomy: Inferring the Similarity of Task-Derived Representations from Brain Activity (NeurIPS 2019 · similarity 0.89)
- Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex (NeurIPS 2025 · similarity 0.89)
- In Silico Mapping of Visual Categorical Selectivity Across the Whole Brain (NeurIPS 2025 · similarity 0.87)
- Geometry of naturalistic object representations in recurrent neural network models of working memory (NeurIPS 2024 · similarity 0.87)
- CLIP-MSM: A Multi-Semantic Mapping Brain Representation for Human High-Level Visual Cortex (AAAI 2025 · similarity 0.86)
- BrainLM: A foundation model for brain activity recordings (ICLR 2024 · similarity 0.86)
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.
