"Compressive learning" scaffolds higher-order network structure to enhance human knowledge acquisition

Abstract

Humans naturally seek knowledge, yet integrating vast, fragmented information remains challenging. Traditionally, knowledge acquisition has relied on random walks within network--an unguided and inefficient process. We introduce compressive learning, a framework that embeds higher-order structural features--specifically node-degree inhomogeneity--into pre-learning trajectories to scaffold more efficient learning. Across two large-scale experiments, we demonstrate that scale-free networks--due to their pronounced node-degree inhomogeneity--are more compressible and learnable than other network types, and confirm the efficacy of the compressive learning approach. Magnetoencephalography (MEG) recordings reveal that compressive pre-learning enhances structured neural representations in the dorsal anterior cingulate cortex (ACC). A hypergraph-based two-stage model further reveals that compressive learning constructs a network skeleton of hyperedge-defined substructures that more effectively accommodate new inputs. Together, our results highlight the central role of higher-order network structure in human learning and offer a strategic approach to effectively "connect the dots."