BrainPFN, a transformer meta-trained on synthetic tasks built from real functional connectivity (FC) matrices, predicts behavioural scores for a new cohort in a single forward pass, without refitting. It is most useful in small-sample studies, the common case in fMRI. Related work covers pretraining targets derived from FC and self-supervised learning from raw fMRI time series.

Carlo Alberto Barbano
I am a Starting Researcher in the MIND team at Inria Saclay, working on foundation models for fMRI and brain–behaviour prediction.
Research
An open fMRI collection built from public OpenNeuro studies, uniformly preprocessed and released as parcellated timeseries and FC matrices for several atlases.
- 11,379 subjects
- 45,731 functional runs
- 52 atlases
AnatCL, a foundation model for anatomical brain MRI, normative models for anomaly detection, and models of brain and disease development.
Earlier work covered contrastive learning and debiasing. If you are interested in collaborating, contact me.
Open source software
- BrainPFN Pretrained BrainPFN model for predicting behavioural scores from resting-state FC, with a scikit-learn interface.
pip install brainpfn - AnatCL Code and pre-trained weights for the AnatCL brain MRI models.
pip install anatcl - Simplify Removes the zeroed-out parts of a pruned PyTorch model, so that it actually runs faster.
pip install torch-simplify - torchstain Stain normalization and augmentation for histopathology images, on GPU, with PyTorch, TensorFlow and NumPy backends. I maintain it.
pip install torchstain - cardkit Animated figures with matplotlib: a scene written once in Python, exported as GIF, WebP, MP4 and animated SVG. It powers the animations on this website.
pip install cardkit-animLearn More →
News
- Nov. 2026Open-fMIND is released: parcellated fMRI timeseries and FC matrices from 322 OpenNeuro datasets (11,379 subjects, 52 atlases). Paper · Data.
- Sep. 2026BrainPFN preprint is out.
- May 2026TIMBER preprint, on self-supervised representation learning from raw fMRI time series, is out.
- Nov. 2025I am now a Starting Researcher at Inria Saclay, MIND team.
- Jan. 2025Visiting Researcher at Inria MIND, studying the link between functional connectivity and cognition with contrastive learning.
- Dec. 2024Became a member of the ELLIS Society.
- Mar. 2024Member of the newborn ASL TO3 Radiomics Lab in Rivoli, Italy.
Positions
- Nov. 2025 – nowStarting Researcher, Inria Saclay, MIND
- Jan. – Feb. 2025Visiting Researcher, Inria MIND
- Dec. 2023 – Oct. 2025Postdoctoral Researcher, deep learning for medical imaging, University of Turin
Education
- 2020 – 2023Ph.D. in Computer Science, Télécom Paris and University of Turin (cotutelle). Thesis: Collateral-Free Learning of Deep Representations: From Natural Images to Biomedical Applications
- 2018 – 2020M.Sc. in Artificial Intelligence, University of Turin
Supervision
- Feb. 2026 – nowAkshita Kumar, Ph.D. student, Institut Polytechnique de Paris, co-supervised with Pietro Gori. Self-supervised learning for longitudinal and multimodal brain imaging · Thesis






