AI-guided stimuli discovery and generation to optimize facial emotion perception studies in autism

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Kushin Mukherjee, Na Yeon Kim, Maren Wehrheim, Ralph Adolphs, Kohitij Kar
• Venue: arXiv: cs.AI
• Published: 2026-07-09
• Collected: 2026-07-10
• URL: http://arxiv.org/abs/2607.08533v1
• PDF: https://arxiv.org/pdf/2607.08533v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers have developed a method to create and select facial expressions that enhance differences in emotion perception between autistic and neurotypical adults. This is done by training artificial neural networks on existing data, using the models to pick images predicted to maximize group separation, and synthesizing new images to reduce behavioral separation.

What the Document Says

• The study aims to understand perceptual differences between autistic and neurotypical adults using facial emotion perception as a test case.
• Group differences in emotion judgments have been reported but vary across studies, potentially due to image-level sparsity.
• Artificial neural network models were trained on existing data for autistic and neurotypical participants to predict image-level judgments.
• These models were used to select novel faces predicted to maximize group separation.
• In an independent cohort, model-selected images produced larger behavioral differences than matched random images.
• The same models with a generative adversarial network were used to transform diagnostic images toward greater predicted group agreement.
• Synthesized images reduced behavioral separation relative to their matched originals in validation.
• The research establishes a model-guided framework for discovering and transforming stimuli that reveal population-specific perceptual differences.

Key Concepts Mentioned

• autism
• facial emotion perception
• artificial neural networks
• image-level sparsity
• behavioral assays

Claims Made by the Document

• Group differences in emotion judgments have image-level sparsity.
• Model-selected images produce larger behavioral differences than matched random images.
• Synthesized images reduce behavioral separation.

Evidence Used

• variability in findings across studies
• results from independent cohort and validation

Methods / Approach

• training artificial neural network models on existing data
• using models to select novel faces
• synthesizing new images using generative adversarial networks

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The generalizability of the findings to other populations and types of stimuli is unknown.

Open Questions

• How do these results apply to other neurodivergent conditions or types of perceptual tasks?
• What are the implications for clinical assessment and intervention in autism?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The study focuses on facial emotion perception differences between autistic and neurotypical adults. It suggests that these differences may be concentrated in a small subset of diagnostic facial expressions rather than spread uniformly across stimuli. The researchers use artificial neural network models to select images predicted to maximize group separation and synthesize new images to reduce behavioral separation.

Raw Abstract / Excerpt

Understanding perceptual differences between autistic and neurotypical adults requires behavioral assays that are sensitive, reliable, and mechanistically informative. Facial emotion perception is a useful test case because group differences have been reported, but findings vary across studies. Here we show that this variability may reflect image-level sparsity: autistic-neurotypical differences in emotion judgments were concentrated in a small subset of diagnostic facial expressions rather than spread uniformly across stimuli. We trained population-specific artificial neural network models to predict image-level judgments for autistic and neurotypical participants, then used these models to select novel faces predicted to maximize group separation. In an independent cohort, model-selected images produced larger behavioral differences than matched random images. We then used the same models with a generative adversarial network to transform diagnostic images toward greater predicted group agreement. In phenotype-matched validation, synthesized images reduced behavioral separation relative to their matched originals. These results establish a model-guided framework for discovering and transforming stimuli that reveal population-specific perceptual differences. More broadly, they show 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.