0
1

[–] saprian 0 points 1 point (+1|-0) ago 

Note that the data set is only ~1800 images; not sure this model amounts to much of anything meaningful.

0
1

[–] Tancred 0 points 1 point (+1|-0) ago 

Criminality supposedly has a strong link to intelligence (peaking around IQ 80).
Anecdotally, you can often see if someone is several SDs below average in IQ by just looking at a person's face. (A study about that would be neat)
Makes sense that causal links between criminality and facial structure can be found.

Asian faces compared to Caucasian faces supposedly differ more in the arrangement of facial features rather than the phenotype of individual features, so they'd probably need to train different nets for each race and aggregate the results for mixed races.
Maybe also stay away from transfer learning as the mid level features might differ strongly between races.

0
1

[–] Antiracist7 0 points 1 point (+1|-0) ago 

So deep learning can recognize black people?

@eagleshigh