Deep::DirectKPPCNetwork Class Reference

Predictive Coding network built from DirectKPPCLayer, using Direct Kolen-Pollack feedback alignment (DFA) instead of symmetric weight transport. See the file-level warning above on its current verification status. More...

#include <DirectKPPCNetwork.h>

Public Member Functions
	DirectKPPCNetwork (int batchSize, DeviceType device=DeviceType::DEVICE_CPU) noexcept
	Constructs an empty network with a predetermined batch size.
	~DirectKPPCNetwork ()=default
	Default destructor.
	DirectKPPCNetwork (const DirectKPPCNetwork &)=delete
	Copy construction is disabled; a network owns its layers' memory arena and is not meant to be duplicated.
DirectKPPCNetwork &	operator= (const DirectKPPCNetwork &)=delete
	Copy assignment is disabled; see the copy constructor.
void	AddLayer (size_t size, size_t nextSize, size_t terminalSize, float lr, float ir, float fl, float lmbda, ActivationType aType, ActivationType dType)
	Adds a layer to the network.
void	RandomizeWeights (std::mt19937 &rng)
	Randomizes the weights (W and Psi) of each layer.
void	ResetState () noexcept
	Resets each layer's state without touching learned weights.
void	Clamp (const std::vector< float > &input)
	Clamps the input to the first layer.
void	ProjectForward () noexcept
	Seeds every hidden layer's z from a genuine forward pass through current weights. Assumes layers[0] is already clamped.
float	CalculateTerminalError () noexcept
	Phase 0's final step: computes the terminal layer's error alone, producing a real, non-zero epsilon_L. Call AFTER clamping the target onto the terminal layer, and BEFORE DirectFeedbackUpdate().
void	DirectFeedbackUpdate () noexcept
	Phase 1: runs DirectFeedbackUpdate() on every non-terminal layer. Order among layers does not matter (see class-level docs) – unlike GaussSeidelPCNetwork, no sweep ordering is required here. Still sequential in this revision; see class-level warning.
float	Step (bool computeEnergy=true) noexcept
	Phase 2: runs one ordinary PC settling step (CalculateState() + UpdateState()) across every layer.
void	UpdateWeights () noexcept
	Phase 3: updates every non-terminal layer's W and Psi. Called once after the settling loop completes.
void	SetOptimizer (OptimizerType o) noexcept
	Sets the optimizer used for W updates, on every layer.
void	SetPsiOptimizer (OptimizerType o) noexcept
	Sets the optimizer used for Psi updates, on every layer.
void	SetLearningRate (float lr) noexcept
	Sets the learning rate used for W updates, on every layer.
void	SetFeedbackRate (float fl) noexcept
	Sets the learning rate used for Psi updates, on every layer.
DirectKPPCLayer *	GetTerminalLayer () noexcept
	Returns the network's terminal (final) layer.
std::vector< std::unique_ptr< DirectKPPCLayer > > &	GetLayers () noexcept
	Returns every layer in the network, in the order they were added.
const std::vector< std::unique_ptr< DirectKPPCLayer > > &	GetLayers () const noexcept
	Returns every layer in the network, in the order they were added.
int	GetBatchSize () const noexcept
	Returns the batch size for the network's layers.
DeviceType	GetDevice () const noexcept
	Returns which device this network's layers run on.
float	TrainStep (const std::vector< float > &x, const std::vector< float > &y, int inferenceSteps=1)
	Full train step: reset, clamp input+target, run all four DKP-PC phases in order, unclamp.
std::vector< float >	Predict (const std::vector< float > &x, int inferenceSteps)
	Forward prediction: clamp input, settle with no DFA perturbation and no target clamped, read the terminal's beliefs.
void	Compile ()
	Loads all layers into one contiguous block of memory (MemoryArena for DEVICE_CPU, DeviceMemoryArena for DEVICE_GPU), and wires layerAbove/layerBelow/terminalLayer across every layer.

Detailed Description

Predictive Coding network built from DirectKPPCLayer, using Direct Kolen-Pollack feedback alignment (DFA) instead of symmetric weight transport. See the file-level warning above on its current verification status.

Constructor & Destructor Documentation

DirectKPPCNetwork() [1/2]

Deep::DirectKPPCNetwork::DirectKPPCNetwork ( int batchSize, DeviceType device = DeviceType::DEVICE_CPU )

explicitnoexcept

Constructs an empty network with a predetermined batch size.

Parameters

batchSize	Batch size
device	Which device this network's layers should run on. Defaults to DEVICE_CPU, preserving existing behavior exactly for anyone not explicitly requesting DEVICE_GPU.

~DirectKPPCNetwork()

Deep::DirectKPPCNetwork::~DirectKPPCNetwork ( )

default

Default destructor.

DirectKPPCNetwork() [2/2]

Deep::DirectKPPCNetwork::DirectKPPCNetwork ( const DirectKPPCNetwork & )

delete

Copy construction is disabled; a network owns its layers' memory arena and is not meant to be duplicated.

Member Function Documentation

AddLayer()

void Deep::DirectKPPCNetwork::AddLayer	(	size_t	size,
		size_t	nextSize,
		size_t	terminalSize,
		float	lr,
		float	ir,
		float	fl,
		float	lmbda,
		ActivationType	aType,
		ActivationType	dType )

Adds a layer to the network.

Parameters

size	input size
nextSize	output size (0 marks a terminal layer)
terminalSize	the size of the network's final output layer (e.g. 10 for MNIST) – required on every AddLayer call, not inferred, since the true terminal layer isn't known until the whole network has been assembled. Compile() sanity-checks this against the actual last layer's size.
lr	learning rate for W
ir	inference rate (Euler integration step size)
fl	feedback rate (see Layer implementatio)
lmbda	weight decay (L2 regularization) coefficient, shared between W and Psi
aType	activation type
dType	activation derivative type

CalculateTerminalError()

float Deep::DirectKPPCNetwork::CalculateTerminalError ( )

noexcept

Phase 0's final step: computes the terminal layer's error alone, producing a real, non-zero epsilon_L. Call AFTER clamping the target onto the terminal layer, and BEFORE DirectFeedbackUpdate().

Returns

The terminal layer's energy contribution.

Clamp()

void Deep::DirectKPPCNetwork::Clamp

(

const std::vector< float > &

input

)

Clamps the input to the first layer.

Parameters

input

Reference to the input vector.

Compile()

void Deep::DirectKPPCNetwork::Compile

(

)

Loads all layers into one contiguous block of memory (MemoryArena for DEVICE_CPU, DeviceMemoryArena for DEVICE_GPU), and wires layerAbove/layerBelow/terminalLayer across every layer.

DirectFeedbackUpdate()

void Deep::DirectKPPCNetwork::DirectFeedbackUpdate ( )

noexcept

Phase 1: runs DirectFeedbackUpdate() on every non-terminal layer. Order among layers does not matter (see class-level docs) – unlike GaussSeidelPCNetwork, no sweep ordering is required here. Still sequential in this revision; see class-level warning.

GetBatchSize()

int Deep::DirectKPPCNetwork::GetBatchSize ( ) const

inlinenoexcept

Returns the batch size for the network's layers.

GetDevice()

DeviceType Deep::DirectKPPCNetwork::GetDevice ( ) const

inlinenoexcept

Returns which device this network's layers run on.

GetLayers() [1/2]

const std::vector< std::unique_ptr< DirectKPPCLayer > > & Deep::DirectKPPCNetwork::GetLayers ( ) const

inlinenoexcept

Returns every layer in the network, in the order they were added.

GetLayers() [2/2]

std::vector< std::unique_ptr< DirectKPPCLayer > > & Deep::DirectKPPCNetwork::GetLayers ( )

inlinenoexcept

Returns every layer in the network, in the order they were added.

GetTerminalLayer()

DirectKPPCLayer * Deep::DirectKPPCNetwork::GetTerminalLayer ( )

inlinenoexcept

Returns the network's terminal (final) layer.

operator=()

DirectKPPCNetwork & Deep::DirectKPPCNetwork::operator= ( const DirectKPPCNetwork & )

delete

Copy assignment is disabled; see the copy constructor.

Predict()

std::vector< float > Deep::DirectKPPCNetwork::Predict	(	const std::vector< float > &	x,
		int	inferenceSteps )

Forward prediction: clamp input, settle with no DFA perturbation and no target clamped, read the terminal's beliefs.

ProjectForward()

void Deep::DirectKPPCNetwork::ProjectForward ( )

noexcept

Seeds every hidden layer's z from a genuine forward pass through current weights. Assumes layers[0] is already clamped.

RandomizeWeights()

void Deep::DirectKPPCNetwork::RandomizeWeights

(

std::mt19937 &

rng

)

Randomizes the weights (W and Psi) of each layer.

Parameters

rng	The classic Mersenne Twister

ResetState()

void Deep::DirectKPPCNetwork::ResetState ( )

noexcept

Resets each layer's state without touching learned weights.

SetFeedbackRate()

void Deep::DirectKPPCNetwork::SetFeedbackRate ( float fl )

inlinenoexcept

Sets the learning rate used for Psi updates, on every layer.

SetLearningRate()

void Deep::DirectKPPCNetwork::SetLearningRate ( float lr )

inlinenoexcept

Sets the learning rate used for W updates, on every layer.

SetOptimizer()

void Deep::DirectKPPCNetwork::SetOptimizer ( OptimizerType o )

inlinenoexcept

Sets the optimizer used for W updates, on every layer.

SetPsiOptimizer()

void Deep::DirectKPPCNetwork::SetPsiOptimizer ( OptimizerType o )

inlinenoexcept

Sets the optimizer used for Psi updates, on every layer.

Step()

float Deep::DirectKPPCNetwork::Step ( bool computeEnergy = true )

noexcept

Phase 2: runs one ordinary PC settling step (CalculateState() + UpdateState()) across every layer.

Parameters

computeEnergy

asks for energy to be returned

Returns

Total energy, summed from every layer's CalculateState().

TrainStep()

float Deep::DirectKPPCNetwork::TrainStep	(	const std::vector< float > &	x,
		const std::vector< float > &	y,
		int	inferenceSteps = 1 )

Full train step: reset, clamp input+target, run all four DKP-PC phases in order, unclamp.

Parameters

x	Flattened input batch, clamped to the input layer.
y	Flattened target batch, clamped to the terminal layer.
inferenceSteps	Number of settling steps for phase 2. Defaults to 1, matching the paper's own headline result – unlike this codebase's other PC variants, DKP-PC's entire point is that a single step is enough BECAUSE of the DFA phase, so defaulting to a larger value here would silently waste most of the intended speed advantage.

Returns

The final settling step's total energy, before weight updates.

UpdateWeights()

void Deep::DirectKPPCNetwork::UpdateWeights ( )

noexcept

Phase 3: updates every non-terminal layer's W and Psi. Called once after the settling loop completes.

---

The documentation for this class was generated from the following files:

• include/deepity/networks/DirectKPPCNetwork.h
• src/networks/DirectKPPCNetwork.cpp