Interested in category selectivity and topographic modelling? Come see my poster tomorrow at CCN (A57). We show that encoding models confirm dissociable selective responses to bodies, hands, and tools, and test if topographic ANNs capture that organization.
See you there!

Once again, DANNs could not capture visual cortex action-related organization. Topographic constraints or even training objective (object vs. action classification) did not have an effect on the results.

In a series of successive multivariate analyses, we also show that: 1) action is dissociable not only from animacy but also from shape; 2) the "effector" properties of objects (that characterize hands and tools) is the most relevant action-related dimension shaping LOTC object space.

🧵 11/n

While TDANNs successfully replicated the animacy division seen in VOTC, they completely failed to develop the action-related topography of LOTC.
🧵 10/n

The opposite activation patterns were nicely summarised by our vector-of-ROIs analysis, where we consider a broad chunk of cortex in ventral and lateral OTC.

🧵 8/n

Key Finding: We discovered that LOTC contains a topographic map where objects are organized by action-related features. Activations for bodies, hands, tools, and manipulable objects overlap along a dorsal-posterior to ventral-anterior axis, converging for object effectors (hands and tools).

🧵 6/n

We used fMRI to map brain responses to images with varying action properties:
✋ Animate: Faces, Whole-Bodies, Hands
✂️ Inanimate: Tools (graspable + effector), Manipulable objects (only graspable, not effector), and Non-manipulable objects.

🧵 5/n