Meet Our Current Scholars

Marko seeks to better understand the prioritized attentional processing that the human visual system affords to stimuli that are threatening (e.g., snakes, guns).  He is also exploring the “low prevalence effect,” which is the finding that (during visual search) items that are rare are missed disproportionately often relative to items that are more common.  This project began with the question “Does the prioritization of threatening items afford protection against prevalence effects?”  To explore this question, he had people conduct visual search looking for threatening items (snakes) or nonthreatening items (rabbits).  After finding unexpected results (i.e., threats were actually found less often and more slowly), he replicated those findings and added two conditions: search targets that look similar to snakes but are nonthreatening (caterpillars) and search targets that elicit emotion (disgust) but not fear (roaches).

Laboratory visual search tasks seldom capture the dynamics or complexity of real-world visual search (e.g., medical image or baggage screening). Crystal is working on a project to measure the similarity of scenes within the Oddity Detection in Diverse Scenes (ODDS) database, which is a collection of real-world images/scenes with subtle image deformations/anomalies that may facilitate the study of visual search training protocols. Quantifying the similarity between anomalies can help researchers design experiments and create custom training protocols tailored to the search performance of each individual. The pairwise similarity ratings on a subset of the images from Crystal's human participants will be used to train an AI to rate the pairwise similarity between the remaining anomalies, which is a task that would be far too time-intensive for human participants.

Medical image analysis requires detection of ill-specified deformities, a skill that requires considerable training. After medical school/residency, radiologists must retain their skills and transfer them across screenings. Participants in this semi-longitudinal study used one of four perceptual learning methods to train anomaly detection during weeks 1-3. Following a 2-week break, participants completed a skill retention test and then a transfer task. Training was assessed via performance on simple search arrays and complex scenes.  Performance enhancements were near-universally retained over time, and few performance decrements were observed on altered stimuli, indicating that anomaly detection skills can be retained over time.

Detecting anomalies within medical images is critical to routine/emergency medicine. Here, we used a semi-longitudinal approach to investigate four perceptual learning methods for training a laboratory analogue of medical image analysis. In weeks 1-3, participants developed skills using one of the learning methods. Weeks 4-5 were “off weeks,” followed by a skill retention test in week 6 and a skills transfer task in week 7. Training was assessed via performance on simple search arrays and complex scenes.  Performance increased steadily over time, for both target detection and exhaustive searches, indicating that anomaly detection can effectively be trained via practice.

​