Researchers developed a model-guided framework for discovering and transforming stimuli to reveal population-specific perceptual differences between autistic and neurotypical adults. By training artificial neural network models on image-level judgments, they identified that diagnostic facial expression differences are concentrated in a small subset of stimuli rather than being uniform.
- The team used these models to select novel faces predicted to maximize group separation, which produced larger behavioral differences than matched random images in an independent cohort.
- They employed a generative adversarial network to transform diagnostic images toward greater predicted group agreement, successfully reducing behavioral separation in phenotype-matched validation.
This approach demonstrates how behavioral phenotyping can move beyond averaging across fixed stimulus sets toward optimized assays that identify the conditions under which neurodivergent perception diverges or converges.