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

An Optimal Compression Principle for Event Segmentation and Encoding

Isaac Ashkenazi¹, Yuval Hart¹; ¹Hebrew University of Jerusalem

Presenter: Isaac Ashkenazi

Humans spontaneously segment continuous experience into events, which shape memory and learning. Yet it remains unclear what objective or principle makes particular moments function as boundaries. We propose an information-theoretical framework, Minimum Description Length (MDL), that accounts for behavioral boundaries in sequences of stimuli by balancing global coding simplicity against within-event accuracy. This principled approach optimizes compression to extract generalizable event schema from noisy experiences, producing a family of hierarchical segmentations controlled by one granularity parameter. Across naturalistic video and audio stimuli, MDL boundaries align with human segmentation judgments. In simulations, the same principle recovers latent event structure without hand-specified state priors; and captures the U-shaped curve of free recall. The MDL framework suggests that event boundaries may emerge from general pressure toward efficient representation.

Topic Area: Decision-Making, Cognitive Control & Event Cognition