Hippocampal Memory Enables Continual Learning Without Replay

This method implements a training-only hippocampal memory that indexes input geometry and limits synaptic movement per direction to prevent catastrophic forgetting. It allows fully trainable networks to learn new tasks without relying on data replay, distillation, or frozen backbones.