How does curriculum learning influence cognitive map formation?

Liang Z
Summerfield C

How does learning curriculum shape cognitive map formation? We manipulated the order in which participants learned one-step transitions in two maps, then tested multi-step inference. Grouped training- learning along single rows or columns- fragmented knowledge of the map. By contrast, disjoint training- learning spatially separated transitions- improved inference performance, suggesting a more integrated cognitive map.

Scientific Abstract

Representing relational knowledge in a mental map is useful for adaptive behaviour. Here, we studied how curriculum learning (the use of different training regimes) influences the way that mental (or ‘cognitive’) maps are formed. We investigated how the order in which people learned one-step transitions impacted their subsequent ability to perform multi-step inference using the cognitive maps they had formed. Participants learned one map with a curriculum that grouped information by rows and columns (‘grouped curriculum’) and another that selected transitions for training in a spatially disjoint way (‘disjoint curriculum’). Subsequent multi-step navigation was performed better after training using the disjoint curriculum. We hypothesise that learning under a grouped curriculum results in mental fragmentation of the learned maps, which impairs subsequent navigation across rows and columns. By contrast, learning one-step transitions in a pseudo-random order facilitates subsequent integration of learned information into a mental map to enable adaptive behaviour.

Similar content

Preprint
Causse AA, Curot J, Lopes-Dos-Santos V, Nunes-da-Silva R, Barron HC, Dornier V, Denuelle M, De Barros A, Sol J, Lotterie J, Lehongre K, Fernandez-Vidal S, Frazzini V, Navarro V, Valton L, Barbeau EJ, Denison T, Reddy L, Dupret D

A learning-evoked slow-oscillatory architecture paces population activity for offline reactivation across the human medial temporal lobe

Preprint
Nandi T, Puonti O, Clarke WT, Nettekoven CR, Barron HC, Kolasinski J, Hanayik T, Hinson EL, Berrington A, Bachtiar V, Johnstone A, Winkler AM, Thielscher A, Johansen-Berg H, Stagg CJ

tDCS induced GABA change is associated with the simulated electric field in M1, an effect mediated by grey matter volume in the MRS voxel

How does curriculum learning influence cognitive map formation?

Liang Z
Summerfield C

How does learning curriculum shape cognitive map formation? We manipulated the order in which participants learned one-step transitions in two maps, then tested multi-step inference. Grouped training- learning along single rows or columns- fragmented knowledge of the map. By contrast, disjoint training- learning spatially separated transitions- improved inference performance, suggesting a more integrated cognitive map.

Scientific Abstract

Representing relational knowledge in a mental map is useful for adaptive behaviour. Here, we studied how curriculum learning (the use of different training regimes) influences the way that mental (or ‘cognitive’) maps are formed. We investigated how the order in which people learned one-step transitions impacted their subsequent ability to perform multi-step inference using the cognitive maps they had formed. Participants learned one map with a curriculum that grouped information by rows and columns (‘grouped curriculum’) and another that selected transitions for training in a spatially disjoint way (‘disjoint curriculum’). Subsequent multi-step navigation was performed better after training using the disjoint curriculum. We hypothesise that learning under a grouped curriculum results in mental fragmentation of the learned maps, which impairs subsequent navigation across rows and columns. By contrast, learning one-step transitions in a pseudo-random order facilitates subsequent integration of learned information into a mental map to enable adaptive behaviour.

DOI

10.7554/eLife.111035.1

Downloads

View PDF (2MB)

Similar content

Preprint
Causse AA, Curot J, Lopes-Dos-Santos V, Nunes-da-Silva R, Barron HC, Dornier V, Denuelle M, De Barros A, Sol J, Lotterie J, Lehongre K, Fernandez-Vidal S, Frazzini V, Navarro V, Valton L, Barbeau EJ, Denison T, Reddy L, Dupret D

A learning-evoked slow-oscillatory architecture paces population activity for offline reactivation across the human medial temporal lobe

Preprint
Nandi T, Puonti O, Clarke WT, Nettekoven CR, Barron HC, Kolasinski J, Hanayik T, Hinson EL, Berrington A, Bachtiar V, Johnstone A, Winkler AM, Thielscher A, Johansen-Berg H, Stagg CJ

tDCS induced GABA change is associated with the simulated electric field in M1, an effect mediated by grey matter volume in the MRS voxel