[x] Re-render adult drawings from all 38 classes now that we have enough kid drawings in all classes. Maybe when re-extracting features, let's collapse across spatial dimension, resulting in vectors of length #channels for each layer?
[x] Reproduce analyses from CogSci paper with "new" sketches
[x] Resolve adult noise-ceiling puzzle -- which consisted of adult-adult correlations being lower in some layers than even olderkids-adult?
[x] Estimate kid vs. adult variance within-class. Compare with other possibility, which is that the layout of centroids is actually different (measured by point estimate RDM)
[x] Classifier performance when training on young, testing on young vs. testing on old and vice versa?
For museum station: (possibly also bing station / jmz station!)
[x] Design a motor task: e.g., tracing a star or other interesting abstract shapes
Save for later:
[x] How well do kids recognize each other's drawings, according to age category?
[x] Collecting more recognition data on the new sketches/categories -- talked to Mike, he suggested not totally worth it for cogsci paper since we have bigger goals (i.e., followup studies)
[x] Are the classes that are most recognizable to adults also those best recognized by model? Look at confusion matrix for humans and model
For museum station: (possibly also bing station / jmz station!)
Save for later: