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

Human inference reflects distinct, interacting contributions of information compression and computational complexity

Jacob A Parker¹, Kristen Li¹, Vijay Balasubramanian¹, Joseph Kable¹, Joshua Gold¹; ¹University of Pennsylvania

Presenter: Jacob A Parker

Human inference is suboptimal for many tasks. One potential cause of this suboptimality is that optimal inference is costly to implement, leading people to reduce the complexity of their strategies. Although many notions of complexity have been used to study this idea, how they relate to each other theoretically and with respect to behavior remains unclear. We developed and used new theoretical and empirical approaches to investigate the role of complexity in human inference. We identified a distinction between two commonly considered forms of complexity: information compression (i.e., an "information bottleneck" on inference) and computational complexity (related to the mathematical operations used to perform inference). Variation in these two forms of complexity explained almost all individual variability in performance on a classic inference task. Moreover, when we increased the computational complexity of optimal inference, participants using optimal strategies more substantially compressed sensory evidence than those opting for a computationally simple strategy. These results shed light on the nature of different forms of complexity and reveal their distinct but interacting contributions to human inference.

Topic Area: Methods, Tools, Theory & Neural Coding