Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

Human-in-the-loop synthesis of behaviorally salient “super-distractors”

Debadrita Sen¹, Achin Parashar², Abhimanyu Ray², Shristi Chourasiya², Devarajan Sridharan³; ¹IISER Berhampur, ²Indian Institute of Science, Indian institute of science, Bangalore, ³Indian Institute of Science, Bangalore (IISc)

Presenter: Devarajan Sridharan

Working memory(WM) is the ability to maintain and manipulate information that is no longer present in the environment. The resilience of WM to distraction is largely tested by studies employing simple stimuli (e.g., gratings, shapes, isolated objects). Hence, what kinds of complex, naturalistic images make for potent WM distractors remains unknown. Here we leverage recent advances in deep generative models to synthesize naturalistic images that powerfully disrupt WM. Our approach generates synthetic images with a class-conditional generative adversarial network (GAN), while concurrently testing the efficacy of these images as distractors on participants(n=16) performing a spatial WM task (human-in-the-loop). With a genetic algorithm for optimization, we identify the most salient feature combinations and refine them over generations to produce powerful “super-distractor” images. Our study demonstrates the feasibility of generating novel kinds of images optimized for specific behaviors, with a human-in-the-loop paradigm.

Topic Area: Memory, Spatial Cognition & Skill Learning

Extended Abstract: Full Text PDF