Class GraphNode<T>

java.lang.Object

org.episteme.core.mathematics.ml.neural.autograd.GraphNode<T>

All Implemented Interfaces:

Serializable

public class GraphNode<T>
extends Object
implements Serializable

Represents a node in the computation graph for Automatic Differentiation.

Nodes wrap Tensors and track the operation history to allow for reverse-mode automatic differentiation (backpropagation).

Since:

2.0

Author:

Silvere Martin-Michiellot, Gemini AI (Google DeepMind)

See Also:

• Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static interface	GraphNode.BackwardFunc<T>	Functional interface for the backward pass of an operation.

Constructor Summary

Constructors

Modifier	Constructor	Description
	GraphNode(Tensor<T> data)
	GraphNode(Tensor<T> data, boolean requiresGrad)
protected	GraphNode(Tensor<T> data, boolean requiresGrad, GraphNode.BackwardFunc<T> backwardFunc, List<GraphNode<T>> parents, String op)

Method Summary

Modifier and Type	Method	Description
GraphNode<T>	add(GraphNode<T> other)	Element-wise addition.
void	backward()	Triggers reverse-mode automatic differentiation.
GraphNode<T>	broadcast(int... shape)	Broadcasts the node to a new shape.
Tensor<T>	getData()
Tensor<T>	getGrad()
GraphNode<T>	log()	Natural logarithm.
GraphNode<T>	matmul(GraphNode<T> other)	Matrix multiplication (supports multidimensional tensors via einsum).
GraphNode<T>	mean()	Arithmetic mean of all elements.
GraphNode<T>	multiply(GraphNode<T> other)	Element-wise multiplication (Hadamard product).
GraphNode<T>	negate()	Element-wise negation.
GraphNode<T>	relu()	Rectified Linear Unit (ReLU).
boolean	requiresGrad()
GraphNode<T>	scale(double factor)	Scale by a scalar factor.
void	setData(Tensor<T> data)
void	setGrad(Tensor<T> grad)
GraphNode<T>	sigmoid()	Sigmoid activation function.
GraphNode<T>	subtract(GraphNode<T> other)	Element-wise subtraction.
GraphNode<T>	sum()	Sum of all elements in the tensor.
String	toString()

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Details

GraphNode

public GraphNode(Tensor<T> data)

GraphNode

public GraphNode(Tensor<T> data,
 boolean requiresGrad)

GraphNode

protected GraphNode(Tensor<T> data,
 boolean requiresGrad,
 GraphNode.BackwardFunc<T> backwardFunc,
 List<GraphNode<T>> parents,
 String op)

Method Details

backward

public void backward()

Triggers reverse-mode automatic differentiation. Computes the gradient of this node with respect to all its ancestors.

getData

public Tensor<T> getData()

getGrad

public Tensor<T> getGrad()

setGrad

public void setGrad(Tensor<T> grad)

requiresGrad

public boolean requiresGrad()

add

public GraphNode<T> add(GraphNode<T> other)

Element-wise addition.

multiply

public GraphNode<T> multiply(GraphNode<T> other)

Element-wise multiplication (Hadamard product).

matmul

public GraphNode<T> matmul(GraphNode<T> other)

Matrix multiplication (supports multidimensional tensors via einsum).

negate

public GraphNode<T> negate()

Element-wise negation.

subtract

public GraphNode<T> subtract(GraphNode<T> other)

Element-wise subtraction.

relu

public GraphNode<T> relu()

Rectified Linear Unit (ReLU).

sigmoid

public GraphNode<T> sigmoid()

Sigmoid activation function.

log

public GraphNode<T> log()

Natural logarithm.

sum

public GraphNode<T> sum()

Sum of all elements in the tensor.

scale

public GraphNode<T> scale(double factor)

Scale by a scalar factor.

mean

public GraphNode<T> mean()

Arithmetic mean of all elements.

setData

public void setData(Tensor<T> data)

broadcast

public GraphNode<T> broadcast(int... shape)

Broadcasts the node to a new shape.

toString

public String toString()

Overrides:

toString in class Object