Dynamic Field Theory

Dynamic Field Theory models activation as a continuous state over a metric space. A neural field receives input, relaxes toward a resting level, interacts laterally with nearby positions, and produces an output through a transfer function.

For a field activation u, a common form is:

tau * du/dt = -u + h + input + lateral_interaction + global_inhibition + noise
output = sigmoid(u)

JUNIPER implements this as fixed-step simulation with JAX arrays.

Step	Purpose
NeuralField	Dynamic activation field.
SpaceToRateCode	Convert a spatial peak to a compact value.
RateToSpaceCode	Convert a compact value to a spatial activation pattern.
HebbianConnection	Learned associative connection.
BCMConnection	BCM-style learned connection.