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

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

It’s a feature, not a bug: Multi-granular world models explain inattentional blindness

Mario Belledonne¹, Ilker Yildirim¹; ¹Yale University

Presenter: Mario Belledonne

Why do we sometimes miss what’s right in front of us? Does our sometimes striking inability to notice a “gorilla” strolling in our midst repudiate the computational sophistication of human vision? Instead of regarding this and the related phenomena of inattentional blindess (IB) as a human quirk, here, we posit the opposite: that it is a signature of several advantageous computational adaptations. We realize this hypothesis by developing goal-conditioned world models, the first-ever model that reverse-engineers human IB by precisely capturing the elements relevant to our goals, while coarsely summarizing the rest scene.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF