Figure 1. Schematic illustration of dynamic continuous attractors as a basis of different neural processes.

Top row: a scheme of continuous attractive manifolds, with a dynamic component in 1D (a), 2D (b) and 3D (c). The neural activity quickly converges on the attractive manifold (dotted arrows), then slides along it (full arrows), producing a dynamics that is temporally structured and constrained to a low dimensional subspace. Bottom row: multiple dynamic memories could be useful for route planning (top left), involved in mind wandering activity (bottom left) or represent multiple learned motor programs (right).