Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Learning what matters: Discovering reward-indicative dimensions from naturalistic objects

Yifan Luo¹, Andrea I Costantino², Hans Op de Beeck², Kobe Desender²; ¹Katholieke Universiteit Leuven, ²KU Leuven

Presenter: Yifan Luo

In the naturalistic environment, how do humans discover which properties of naturalistic objects predict reward? Objects possess structure across dozens of dimensions simultaneously, and which dimensions matter for reward is unknown to the learner. We introduce an object collection paradigm using stimuli from the THINGS dataset, where reward is determined by a single latent dimension that participants must infer through trial-and-error feedback. Pilot data (N=12) demonstrate that humans successfully discover reward-indicative dimensions in this high-dimensional space: we observe progressive improvement in decision value, increasing sensitivity to the reward dimension, and successful generalization to novel objects during a post-learning arrangement task. Preliminary evidence suggests that a dimension's contribution to the global similarity structure modulates learning efficiency. These findings provide initial behavioural evidence that humans can extract task-relevant structure from naturalistic object representations and use it to guide future decisions.

Topic Area: Memory, Learning & Knowledge Structures