Sign in

iiinova is part of the Atmosphere — an open, public network where you own your identity.

David Watson

@dmwatson.bsky.social
534 followers 226 following 41 posts

#PlaceLab

PostsRepliesMedia
David Watson @dmwatson.bsky.social · 26/05/2026
Importantly, we find good correspondence of the components between the MEG, EEG, and fMRI datasets. This indicates the PLSR identified robust neural representations which are at least partially shared over subjects, imaging modalities, and experimental paradigms.
110 1
David Watson @dmwatson.bsky.social · 26/05/2026
Component scores indicate many object features are represented on each component (seen here for MEG again). Each component embodies multiple features, suggesting the brain is tuned to statistical co-occurrences of features amongst natural variation in real world objects.
100 1
David Watson @dmwatson.bsky.social · 26/05/2026
Similarly, the component loadings show smooth oscillations post ~100ms, along with smooth spatial topographies over sensors, as they track the evoked response over time (seen here for an example MEG subject)
100 1
David Watson @dmwatson.bsky.social · 26/05/2026
This model accurately predicts the neural response from ~100 ms onwards (seen here for the MEG dataset)
100 1
David Watson @dmwatson.bsky.social · 26/05/2026
We use Partial Least Squares Regression to derive 5 latent components predicting evoked neural responses over space & time from a set of stimulus dimensions obtained from a previous data-driven analysis of behavioural similarity judgements (www.nature.com/articles/s41...)
100 1
David Watson @dmwatson.bsky.social · 02/03/2026
We also compare connectivity fingerprints for individual regions in the face & scene networks. Fingerprints are similar but still distinct between regions in each network, indicating connectivity profiles track differences in functional specialisations between regions.
Correlations of region-level connectivity fingerprints. Fingerprints are strongly correlated between regions in the face and scene networks, but even more strongly correlated within regions.
100 1
David Watson @dmwatson.bsky.social · 02/03/2026
Connectivity fingerprints are highly similar across people, but also more similar within a person than between people, showing connectivity profiles of face & scene regions track individual differences.
Correlations of connectivity fingerprints within and between subjects. Connectivity fingerprints are strongly correlated between subjects, but even more strongly correlated within subjects.
100 1
David Watson @dmwatson.bsky.social · 02/03/2026
We measure whole-brain functional connectivity "fingerprints" of face & scene regions, representing patterns of connectivity with the rest of the brain, in multiple movie-watching and resting-state datasets. We compare these fingerprints across 2 independent splits of each dataset.
Connectivity fingerprints of face and scene regions with the rest of the brain. Fingerprints are distinct between different regions, and consistent across data splits.
121 2
David Watson @dmwatson.bsky.social · 13/02/2025
So, a fairly emphatic null result. Many connections we wouldn't expect to correlate with behaviour, but we don't even see associations for connections with dorsal attention, default mode, or frontoparietal regions which the scene regions show connectivity biases towards. Why not?
100 1
David Watson @dmwatson.bsky.social · 13/02/2025
BUT, how about the connectivity with the subcortical structures? Also no.
Correlations of functional connectivity between scene regions & subcortical structures with performance on the behavioural measures.
100 1
David Watson @dmwatson.bsky.social · 13/02/2025
Many of the connectivity biases were with cortical networks throughout the rest of the brain - do these correlate any better with behaviour? Still no. Again, we get a mix of small positive and negative correlations - no clear association between connectivity and behaviour.
Correlations of functional connectivity between scene regions & cortical networks with performance on the behavioural measures.
100 1
David Watson @dmwatson.bsky.social · 13/02/2025
Turns out no. For connectivity between our main ROIs, we just get a mix of small positive and negative correlations with behaviour, most of which aren't significant.
Correlations of functional connectivity between main ROIs with performance on the behavioural measures.
100 1
David Watson @dmwatson.bsky.social · 13/02/2025
We measured resting-state functional connectivity of scene regions to other ROIs, and to cortical networks & subcortical structures throughout the brain. As before, OPA connects more with early visual & dorsal attention networks, RSC more with default mode & frontoparietal networks & hippocampus.
Group average resting-state functional connectivity of scene regions (OPA, PPA, RSC) with other pre-selected regions of interest, and with cortical networks and subcortical structures throughout the rest of the brain.
100 1
David Watson @dmwatson.bsky.social · 13/02/2025
We selected behavioural measures of scene recognition, spatial memory, and navigational ability - there was substantial individual variation in performance on all tasks.
Distributions of scores on recognition memory for scenes, the Santa Barbara Sense of Direction Scale, the spatial subscale of the Survey of Autobiographical Memory, and the real-world Navigation test. All tests indicate extensive individual variation in performance.
100 1
David Watson @dmwatson.bsky.social · 17/01/2025
Nice to see the Trends in Neurosciences contents alert today includes a correction to a paper titled "Where is the error?". Turns out the answer is Box 2.
040 4
David Watson @dmwatson.bsky.social · 13/01/2025
Component scores show many object features varying along components - including animacy, shape, size, and category. But representations do not straightforwardly align with previous accounts - many components embodied multiple features, and many features were present in multiple components.
Top and bottom 5 objects scored along each component. #1 includes tangible objects vs textured images. #2 includes man-made vs natural objects. #3 includes size, shape, and animacy. #4 includes size, but also abstract features such as arousal or excitement. #5 includes shape, but also abstract features such as value or pleasantness.
100 1
David Watson @dmwatson.bsky.social · 13/01/2025
Component loadings indicate continuous graded representations of each component across the visual brain. #1 runs anterior-posterior, #2 runs medial-lateral, and further components variously identify repeating medial-lateral and anterior-posterior gradients.
Component loadings plotted on cortical surface. Loadings indicate maps showing continuous graded representations across the visual brain.
100 1
David Watson @dmwatson.bsky.social · 13/01/2025
Despite reduction in dimensionality (from 66 stimulus features down to just 5 components), PLSR accurately predicts neural responses to objects held out from model fitting throughout ventral visual cortex, as well some lateral occipital and early visual regions.
Prediction accuracies for objects held out from model fitting, plotted on cortical surface. High accuracies in ventral visual stream, lateral occipital cortex, and early visual regions.
100 1
David Watson @dmwatson.bsky.social · 13/01/2025
We use Partial Least Squares Regression to predict brain responses from the stimulus features while also reducing dimensionality. We identify a small number (5) of latent components which capture a low-dimensional representation of the object properties and neural responses.
Model schematic. PLSR predicts neural responses from stimulus features of object images while also reducing dimensionality to 5 latent components. Prediction accuracy is measured on objects held out from model fitting.
100 1