GaussSeidelPCNetwork.h File Reference

Orchestrates GaussSeidelPCLayer's three-phase settling step in the correct order – the layer class alone cannot produce genuine Gauss-Seidel dynamics; the SWEEP ORDER across layers is what makes it Gauss-Seidel rather than Jacobi, and that only exists at this level. More...

#include <vector>
#include <memory>
#include <random>
#include <deepity/layers/GaussSeidelPCLayer.h>
#include <deepity/utils/MemoryArena.h>
#include <deepity/utils/Activations.h>

Go to the source code of this file.

Classes
class	Deep::GaussSeidelPCNetwork
	Predictive Coding network built from GaussSeidelPCLayer, sweeping layers sequentially rather than updating them simultaneously. See the file-level warning above on its current verification status. More...

Namespaces
namespace	Deep
	This file exists to provide both Tensor.h and IComputeBackend.h with device types without introducing circular dependencies.

Detailed Description

Orchestrates GaussSeidelPCLayer's three-phase settling step in the correct order – the layer class alone cannot produce genuine Gauss-Seidel dynamics; the SWEEP ORDER across layers is what makes it Gauss-Seidel rather than Jacobi, and that only exists at this level.

Per settling step, in order:

1. UpdateState() on EVERY layer (uses PREVIOUS timestep's mu/e)
2. ComputePrediction() on EVERY layer (uses THIS timestep's fresh z)
3. ComputeError() on EVERY layer (uses THIS timestep's fresh z and layerBelow's fresh mu)

Order among layers WITHIN a single sweep does not matter (verified via the dependency analysis in GaussSeidelPCLayer.h's class docs – each layer's step-1/2/3 only depends on ITS OWN state plus a neighbor's value from a sweep that has ALREADY fully completed). Only the ORDER OF THE THREE SWEEPS THEMSELVES is load-bearing.

Warning

Not yet gradient-checked at the network level. The individual layer's math should be independently verified before trusting any accuracy conclusion drawn from real training with this class.