Emergence of a 'visual number sense' in hierarchical generative models

315 citations recorded by the source corpus.

Citations in the corpus

Theory of Mind May Have Spontaneously Emerged in Large Language Models (2023, unknown): “Models trained to process images, for example, spontaneously learned how to count (27, 28) and differentially process central and peripheral image areas (29), as well as experience human-like optical illusions (30).”
Evaluating large language models in theory of mind tasks (2023, unknown): “Models trained to process images, for example, spontaneously learned how to count ( 30 , 31 ) and differentially process central and peripheral image areas ( 32 ), as well as experience human - like optical illusions ( 33 ).”
“Look, some Green Circles!”: Learning to Quantify from Images (2016, unknown): “Stoianov and Zorzi (2012) show that the ANS emerges as a statistical property of images in deep networks that learn a hi-75 erarchical generative model of visual input.”