Vikas N. O’Reilly-Shah & Alessandro Maria Selvitella 

Journal of Computational Neuroscience

Neural population activity in sensory cortex is organized on low-dimensional manifolds, but it is unclear why such manifolds should arise and what determines their geometry. We address this sensory representation problem by modeling cortical populations as recurrent circuits driven by low-dimensional, regular sensory dynamics (e.g. motion on a circle, head direction, multi-frequency tones on tori). By combining tools from generalized synchronization and delay-embedding theory, specialized to this quasiperiodic regime, we show that contracting recurrent networks generically develop smooth internal manifolds that embed the sensory dynamics. The dimensional requirement is modest and depends only on the intrinsic dimension $$\varvec{d}$$ of the effective sensory manifold, not on the complexity of the external world: a hidden dimension $$\varvec{N>2d}$$ generically suffices (e.g. $$\varvec{N\ge 3}$$ for a circle, $$\varvec{N\ge 5}$$ for a two-frequency torus; bounds compatible with Whitney and Takens’ embedding theorems). We then prove a prediction–separation result that links representational geometry directly to predictive performance, without assuming knowledge of contraction rates: if the circuit can predict future sensory inputs with small error, then states with different futures must be separated in neural state space, up to a resolution set by the prediction error. The resulting scale-limited embeddings naturally give rise to categorical boundaries, metameric equivalence of distinct stimuli, and discrimination thresholds. Numerical experiments with trained $$\varvec{\tanh }$$ recurrent networks driven by head-direction-like and multi-frequency signals recover ring- and torus-shaped hidden manifolds with the expected topology; state separation improves most rapidly near the $$\varvec{2d+1}$$ threshold. Training typically pushes the networks beyond the strict contraction regime where the theory guarantees faithful embedding, yet convergence consistent with generalized synchronization and manifold recovery persist, indicating that our conditions are sufficient but not necessary. Together, these results provide a mechanistic account of why low-dimensional sensory manifolds emerge in recurrent circuits and how prediction constrains their resolution, grounded in dynamical systems embedding theory and consistent with empirical findings on cortical population dynamics.

Read the full article at: link.springer.com

John Stewart

An acceptable information-processing theory of consciousness should be able to identify the adaptive advantages that drove the emergence of consciousness during the evolution of life. It should also predict the specific dynamical architecture of information processing that would need to be instantiated in AI to produce consciousness and the superior adaptation it enables. Whether such an instantiation produces AI that is actually conscious and also more adaptable would provide the ultimate test of the theory. A prime candidate for such a theory is the Subject-Object Emergence Theory of consciousness. It argues that consciousness first evolved because it enabled organisms to achieve adaptive body-environment coordination without extensive trial-and-error learning. It postulates that the subject in an appropriate Subject-Object subsystem would be able to use depictive (iconic) visual representations of the relative positions of its body and the environment to guide motor actions that will produce adaptive body-environment coordination. The depictive representations will ‘light up’ for such a subject, producing subjective experience that is used to deliver adaptive benefits. Hand-eye coordination is a familiar example in humans-novel and intricate coordination tasks can be undertaken without additional reinforcement learning, provided focused conscious attention is employed to provide us (the subject) with relevant depictive images. The paper identifies how such a conscious Subject-Object subsystem could be instantiated in AI systems, enabling hand-eye and other body-environment coordination without the extensive reinforcement learning or complex computational programming needed at present. Drawing further on the Subject-Object theory of consciousness, the paper also identifies how these simple conscious subsystems evolved further in organisms to establish the conscious modelling that enables conscious planning, imagining, abduction and other higher cognitive functions. It demonstrates that current approaches to incorporating world modelling in AI will fail to achieve key elements of the general intelligence found in humans that require consciousness.

Read the full article at: papers.ssrn.com