
From the Lab of Thomas Serre
Associate Director of the Center for Computational Brain Science, Director for the Center for Computation and Visualization, Professor of Cognitive, Linguistic and Psychological Sciences, Professor of Computer Science
Associate Director of the Center for Computational Brain Science, Director for the Center for Computation and Visualization, Professor of Cognitive, Linguistic and Psychological Sciences, Professor of Computer Science
Understanding shortcuts and other ways AI systems do their work is critical for checking that work to determine whether its accurate.
AI researchers and even the general public can visit the LENS site and select from over 1,000 object categories – from a bagel to a ballplayer – and see what visual features large AI models use to identify images and see how those models rank the importance of those features in deciding what objects they’re “seeing.” Users can browse the site to spot examples of so-called shortcuts used by the model. For instance, a lot of the bee images are identified because of the flowers present in the image rather than the insect itself.
LENS was rolled out in 2023, and some researchers are using it as a teaching tool as well as in the lab.