6. Lastly, we found reliable results across two different populations from the US and Mexico. We also computed our data across different distance metrics, correlation methods, and statistical methods. so we are looking forward to feedback! To learn more, check out our preprint!😊

5. Finally, we trained CNNs and algorithmic models on our image dataset (with 40 categories) on object-based properties, and we found that only supervised and models with high performance correlate significantly with object-based and neural spaces.

4. When looking at their 2nd-level bimodal correlations, we found that only object-based properties correlate significantly with neural regions (being more strongly correlated with visual areas than associative areas). Our behavioral and bimodal results were supported by MDS.

3. We computed traditional noise ceilings across brain regions, but also added analyses to compute noise ceilings per ROI, which we believe may be a more informative way of contrasting explainable variance across models to show how it relates to bimodal correlations :)

2. Neurally, we evaluated 11 ROIs between visual and pre-frontal regions to compare the granularity of their representational spaces, and show lower across-subjects variance in visual regions compared to prefrontal areas, which is an important factor for bimodal correlations ;)

1. By collecting similarity jugments for multiple object properties (e.g. category, shape), subjective feelings (scariness and preference), and more abstract ratings (general similarity), we identified what we call an "object-based" space and a "subject-based" space.

There’s much more in the paper! Steve and I look forward to your thoughts in the peer commentary that will come with the article (the call for commentaries will come out soon). You can find the paper here: doi.org/10.1017/S014... and preprint here: osf.io/preprints/ps....

3. We advocate for the usage of Noise Ceilings per ROI (rather than the whole group as usual) to understand better behavior-Neural correlations at a 2nd-level.

Main take-home points:
1. There are at least 2 mental representational spaces across properties.
2. Only "object-based" representational spaces & high-performance CNNs correlate significantly with neural ROIs.

Here is our collaborative work with @meganakpeters.bsky.social @vincenttd.bsky.social and @brianodegaard.bsky.social presented at CNS2025 on the comparison between behavioral similarity judgments on different properties, fMRI, and AI/algorithmic Representational spaces. qr.me-qr.com/es/link-list...

A potential issue might be to control for the angular separation in the internal stimuli condition to attribute/assure repulsion or not :)

I know it is not exactly what you want, but I think the paradigm we developed might easily be adapted to evaluate repulsion contrasting internal vs external stimuli to test your ideas, Here I leave our last year VSS poster: docs.google.com/presentation....