During free viewing, humans look most frequently at people and objects critical to understanding scenes.
New lab paper @natcomms.nature.com: Where do humans look during free viewing? Most frequently, to people and objects that are critical to understanding scenes. Congrats
@smurlidaran.bsky.social
www.nature.com/articles/s41...
Left: conceptual figure representing the analysis of the inner workings of AI (CNNs) to guide researchers' search for neuron types in biological brains; Center: Location opponent CNN neuronal units. One example of an emergent "neuronal" unit type in a CNN manifesting covert attention behaviors despite lacking a built-in attention mechanism. The type of neuron has not been highlighted previously in brains as related to covert attention. Right: Re-analyses of mice superior colliculus neuronal data (Wang, Herman, Krauzlis, 2022) showing location opponent and also location summation neurons.
New PNAS paper!
Probing CNNs with neuroscience tools uncovers emergent covert-attention behavior and predicts previously unnoticed target/cue-location–opponent cells, validated in mice superior colliculus. Congrats @sudh8887.bsky.social @williamwangnlp.bsky.social
Read more: shorturl.at/onk6B
Illustration of a radiologist viewing a breast x-ray image, linear model observer, and a convolutional neural networl
New lab paper in J. Med. Imaging explains when and why CNNs outperform traditional linear models in predicting radiologist performance and assessing medical image quality. Congrats, 1st author Aditya!
Full paper: shorturl.at/k84D9
Vision Science meets Radiology. Two studies, which record eye movements, demonstrate how and why AI can enhance radiologists' search efficiency in mammograms. And how the benefits could be higher for 3D Digital Breast Tomosynthesis.
pubs.rsna.org/doi/abs/10.1...
pubmed.ncbi.nlm.nih.gov/38988989/
