iiinova — where your digital self keeps evolving

- GcSS functional localization substantially reduces sample-size requirements compared with traditional group analyses.

7/ Practical implications: - Building a new GcSS localizer? Aim for ~100–200 subjects to derive stable parcels. - Running a new experiment with existing parcels? N ≈ 10–20 may already provide robust fROI effect estimates.

6/ Lastly, We show that using the same localizer contrast does not guarantee functionally homogeneity. Researchers should validate the response profiles of localized fROIs before assigning them a common functional interpretation.

5/ We also observed the same overall scaling behavior in the cerebellum and in surface-based cortex analyses.

4/ Once robust group parcels exist, even N=10–20 is sufficient to obtain accurate population effect-size estimates for most practically meaningful effects.

140

3/ For most contrasts, probabilistic maps exhibit good between-sample consistency by N ≈ 100 and excellent consistency by N ≈ 200. The same is true for group parcels.

2/ First, we showed that subject-specific functional localization matters. Brain activation patterns are much more consistent within individuals than across individuals, and GcSS produces substantially larger effect sizes than traditional group analyses.

1/ We analyzed task-fMRI data from 900+ Human Connectome Project participants, spanning 16 contrasts across language, multiple-demand, default mode, social, motor, and visual systems. We asked how many subjects are needed for each stage of the Group-Constrained Subject-Specific (GcSS) pipeline.

Eyas Ayesh @eyasayesh.bsky.social · 20/07/2026

Reposted by Ruimin Gao 高睿敏

🚨New position piece! “Deception-Enabling Cognitive Capabilities in AI Systems” 🧠🤖 w/ @dobyrahnev.bsky.social @apurvaratan.bsky.social @neuranna.bsky.social To better identify and mitigate deceptive AI behaviors, we need a deeper understanding of their complex and multifaceted nature...🧵👇

1115

Reposted by Ruimin Gao 高睿敏

Anna (Anya) Ivanova @neuranna.bsky.social · 04/06/2026

Now out in Imaging Neuroscience! The language network responds robustly to sentences across tasks doi.org/10.1162/IMAG...

1103

7/ Practical implications: Language localizers using the sentences>nonwords contrast are robust to task variation. But if your localizer includes an active task, ensure the control condition is at least as difficult as the critical one, or you’ll mix language and MD networks.

000

6/ Conclusion: The language network is primarily input-driven. Although modestly modulated by task demands, its response profile and activation pattern remain stable across tasks. Task demands, however, engage the MD network.

100

5/ In the language network, we can reliably decode stimulus (sentences vs. nonwords), and more accurately than we can decode task. In contrast, in the MD network, task is better decoded than stimulus type.

121

4/ Task demands do increase responses in the language network, but reading sentences accompanied by active tasks also strongly recruits the Multiple Demand (MD) network, which is sensitive to task demands.

100

3/ The activations are remarkably consistent within individuals across tasks (and, as reported before, variable across individuals).

100

2/ Across all six tasks, the language network is strongly engaged by the sentences > non-words contrast.

100

1/ We ran six versions of a language localizer, ranging from passive reading to sentiment judgments.

New preprint w/ @evfedorenko.bsky.social, @neuranna.bsky.social , Chandler Cheung, Matthew Siegelman, Alvincé Pongos, @hopekean.bsky.social , Alyx Tanner

121

The language network responds robustly to sentences across diverse tasks

A left frontal-temporal network selectively supports language comprehension and production. Are computations in this language network driven primarily by bottom-up input, or by top-down task demands? 🧵👇 www.biorxiv.org/content/10.6...

biorxiv.org

A network of left frontal and temporal brain areas supports language comprehension and production, implementing computations related to word retrieval and combinatorial linguistic processing. Here, we...

1175

Reposted by Ruimin Gao 高睿敏

Anna (Anya) Ivanova @neuranna.bsky.social · 16/08/2025

Many thanks to the volunteer organizers and the flash talk presenters for making the CCN watch party at Georgia Tech a success! And thanks all attendees for coming and engaging in discussions!

152

Reposted by Ruimin Gao 高睿敏

Anna (Anya) Ivanova @neuranna.bsky.social · 28/07/2025

Looking forward to #CogSci2025 ! Find us throughout the conference

0296

GitHub - alfnie/spm_ss: Subject-specific fMRI analysis toolbox (evlab.mit.edu)

P.S. If you’re a Matlab user, you can try using the spm_ss toolbox developed by Alfonso (which we here adapted for Python+BIDS) github.com/alfnie/spm_ss

github.com

Subject-specific fMRI analysis toolbox (evlab.mit.edu) - alfnie/spm_ss

052

For a detailed demo with code examples—check out our step-by-step guide 👉 funroi.readthedocs.io/en/latest/ex...

Built to be BIDS-compliant, funROI ensures your data is organized & reproducible. 📁

funROI also provides a wrapper for #Nilearn ’s first-level modeling - Easily run GLM analyses with support for event-related & block designs, customizable hemodynamic responses, confound regression, and statistical contrasts.

3 - Effect Estimation: Quantify the strength of neural responses in your fROIs. 4 - Spatial Correlation: Compare within-subject activation patterns across conditions. 5 - Overlap Estimation: Measure spatial overlap between parcels or fROIs.

2 - fROI Definition: Define subject-specific functional ROIs by selecting the top % of active voxels within each parcel (or use fixed voxel counts/p-value thresholds).

Key features include: 1 - Parcel Generation: Create group parcels (brain masks) from individual activation maps with customizable smoothing & thresholds.

funROI leverages subject-specific functional localization to boost the sensitivity & accuracy of your analyses. It is also easy to use.

120