April 14, 2021

Neural state space alignment for magnitude generalization in humans and recurrent networks

How does the brain learn that both cheetahs and space rockets move “fast,” even though animals and vehicles belong in different semantic categories and do not look alike? We found that when comparing the magnitude of stimuli in different contexts, both humans and recurrent networks learned to represent stimuli along parallel number lines, one for each context. These number lines were normalized to denote a general sense of “more” or “less,” irrespective of context, suggesting a new role for normalization in generalization.


 Read More

Leave a Reply

%d bloggers like this: