
“A biologically inspired modular neural control architecture combining evolutionary phylogenetic hyperparameter optimization with lifelong ontogenetic local learning without global backpropagation through time.”
PHYLOS investigates whether mammalian brain topology and localized learning dynamics can replace global backpropagation in continual autonomous control. The system coordinates six specialized neuro-anatomical modules: Sensory Cortex (predictive coding V1→V2→IT), Prefrontal Cortex (working memory slot attention), Hippocampus (kWTA pattern separation with continuous Modern Hopfield CA3 retrieval), Limbic Drives (homeostatic drive reduction), Cerebellum (forward motor error timing), and Basal Ganglia (striatal D1/D2 action gating). Signals are dynamically orchestrated by a Thalamic Neural Router solving entropy-regularized optimal transport via the Sinkhorn-Knopp algorithm, maintaining robust continual learning across dynamic curriculum stages without catastrophic forgetting.
Dynamic signal routing solving an entropy-regularized optimal transport matrix, preventing capacity collapse across cortical modules.
Synapses optimize strictly forward-in-time via localized Hebbian updates, Kolen-Pollack weight feedback, and predictive coding variational free energy relaxation.
Dentate Gyrus homeostatic kWTA pattern separation coupled with continuous CA3 attractor networks for rapid landmark recall.
Hypothalamic drive-reduction mechanisms (hunger, energy) dynamically modulate executive attention and motivation states.
Explore an interactive WebGL OpenXR spatial asset preview. Rotate, orbit, and toggle mesh visualization modes to inspect simulated sub-millimeter 6DOF coordinate tracking.
