Adaptive Predictive Coding explores a network whose connections can change while it learns. The prototype represents neurons and synapses explicitly and includes updates that can add or prune connections in response to a local confusion signal.
The surrounding experiments include categorization, shape, and room-key tasks. These provide small settings in which to inspect how the network state and connectivity evolve.
The aim is to examine structural adaptation alongside ordinary state updates. It is not evidence that the network reproduces biological learning or discovers an optimal structure. The useful questions are whether structural changes help on a defined task, how stable the resulting behavior is, and whether a fixed-connectivity model achieves the same result.