Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Bridging Brains and Concepts: Interpretable Visual Decoding from fMRI with Semantic Bottlenecks
Authors: Sara Cammarota, Matteo Ferrante, Nicola Toschi
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this study we present an interpretable brain decoding framework that inserts a semantic bottleneck into Brain Diffuser, a well established, simple and linear decoding pipeline. ... Visual brain decoding performance are only slightly lower compared to the original Brain Diffuser metrics (e.g., the CLIP similarity is decreased by 4% for the four subjects), yet offering substantial gains in interpretability and neuroscientific insights. These results show that our interpretable brain decoding pipeline enables voxel-level analysis of semantic representations in the human brain without sacrificing decoding accuracy. |
| Researcher Affiliation | Collaboration | Sara Cammarota Department of Biomedicine and Prevention University of Rome, Tor Vergata Viale Montpellier, 1 Rome (IT) EMAIL Matteo Ferrante Tether Evo Department of Biomedicine and Prevention University of Rome, Tor Vergata EMAIL Nicola Toschi Department of Biomedicine and Prevention University of Rome, Tor Vergata Martinos Center For Biomedical Imaging MGH and Harvard Medical School (USA) EMAIL |
| Pseudocode | No | The paper describes a methodology and framework using textual descriptions and diagrams (Figure 1, Figure 2), but does not contain explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/Sara Cammarota/ Bridging-Brains-and-Concepts. |
| Open Datasets | Yes | Data We used the Natural Scenes Dataset (NSD)[2], a deep f MRI dataset encompassing eight healthy adult subjects who performed a continuous recognition task on thousands of images from the COCO dataset. ... NSD data can be requested at https://naturalscenesdataset.org/. |
| Dataset Splits | Yes | We obtained a training set of 8859 images and 24980 f MRI trials per subject, while the test set included 982 images and 2770 f MRI trials per subject. ... We used 5-fold cross-validation to determine the best values of the regularization coefficient α over logarithmically spaced values in the interval 10 3 α 104. |
| Hardware Specification | Yes | All experiments and model training were carried out on a server equipped with 8 NVIDIA H100 GPUs, 2 TB of RAM, and 256 CPU threads. |
| Software Dependencies | No | The paper mentions: "We use Python library Himalaya Ridge regression with the standard svd solver" and "The analysis was performed in Python using the Nilearn neuro-imaging library [1]." However, it does not specify version numbers for these libraries or Python itself. |
| Experiment Setup | Yes | We used 5-fold cross-validation to determine the best values of the regularization coefficient α over logarithmically spaced values in the interval 10 3 α 104. ... 5-fold cross-validation was used to determine the optimal values of the regularization coefficient α over logarithmically spaced values in the interval 10 6 α 106. ... We fit a ridge regression model to estimate the normalized coordinates... |