This National Science Foundation (NSF) Project Grant award under the Computer and Information Science and Engineering (CFDA 47.070) program will fund research at Brown University to develop computational approaches for aligning deep neural networks (DNNs) with human visual perception strategies. The $1,090,678 award, spanning October 2024 to September 2028, aims to address the growing "misalignment" between the behavior of large-scale DNNs and human visual cognition.
The project will leverage large-scale human visual psychophysics experiments to identify the computational principles underlying object recognition in the human brain. These insights will then be translated into algorithms to train DNNs that better mimic human visual processing. The research will produce a "zoo" of human-aligned DNN variants, and analyze how their neural circuits and representations differ from standard DNNs. This will inform the development of new machine learning approaches, data diets, and objective functions needed to align artificial systems with human visual perception from the outset. The outcomes are expected to yield significant advances in understanding human vision and enable the creation of more human-like artificial intelligence systems.
Generated 3/4/25, 6:19 AM