From Poli et al:
Abstract:
How does complex cognition emerge from simpler underlying processes? We show that two components are sufficient: infant-like curiosity and brain-like biophysical constraints jointly drive the emergence of complex neuronal architectures and cognitive abilities. We first tested curiosity-driven exploration in 275 8-to 15-month-old infants. We then implemented these mechanisms in artificial recurrent neural networks, letting them actively sample 20 cognitive tasks during training, while forcing the networks to operate under brain-like biophysical constraints. These two components were sufficient for a range of biological and cognitive phenomena to emerge: The resulting networks captured human synaptic development, the adult brain’s architecture, and displayed compositional generalisation, a hallmark of human cognition. Curiosity, long viewed as a downstream consequence of complex brains, is also a driver of their complexity.
(This work seems really interesting, but getting a short way into reading the experimental details let me know I am completely unqualified to evaluate it. It is important to note that reproducing selected summary properties does not demonstrate that infant brains develop by the same algorithm. Correlations don't prove causes.)
No comments:
Post a Comment