Class MLPNTMController<T, TInput, TOutput>

Namespace

AiDotNet.MetaLearning.Algorithms

Assembly

AiDotNet.dll

MLP-based NTM controller implementation with learnable parameters.

public class MLPNTMController<T, TInput, TOutput> : INTMController<T>

Type Parameters

T

The numeric type.

TInput

The input data type.

TOutput

The output data type.

Inheritance

object

MLPNTMController<T, TInput, TOutput>

Implements

INTMController<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

This controller uses a multi-layer perceptron to process inputs and generate addressing parameters for the NTM memory. Unlike LSTM, MLP is stateless and processes each timestep independently.

Architecture:

Input (inputSize + numReadHeads * memoryWidth)
  ↓
Hidden Layer 1 (hiddenSize) + ReLU
  ↓
Hidden Layer 2 (hiddenSize) + ReLU
  ↓
Linear projections → ReadKeys, WriteKey, Erase, Add, Output

Constructors

MLPNTMController(NTMOptions<T, TInput, TOutput>)

Initializes a new instance of MLPNTMController with learnable weights.

public MLPNTMController(NTMOptions<T, TInput, TOutput> options)

Parameters

options NTMOptions<T, TInput, TOutput>

The NTM options.

Methods

Forward(Tensor<T>, List<Tensor<T>>)

Forward pass through the controller.

public Tensor<T> Forward(Tensor<T> input, List<Tensor<T>> readContents)

Parameters

input Tensor<T>

The input tensor.

readContents List<Tensor<T>>

The previous read contents.

Returns

Tensor<T>

The controller output.

GenerateAddVector(Tensor<T>)

Generates add vector for writing.

public Tensor<T> GenerateAddVector(Tensor<T> output)

Parameters

output Tensor<T>

The controller output.

Returns

Tensor<T>

The add vector.

GenerateEraseVector(Tensor<T>)

Generates erase vector for writing.

public Tensor<T> GenerateEraseVector(Tensor<T> output)

Parameters

output Tensor<T>

The controller output.

Returns

Tensor<T>

The erase vector.

GenerateOutput(Tensor<T>, List<Tensor<T>>)

Generates final output.

public Tensor<T> GenerateOutput(Tensor<T> output, List<Tensor<T>> readContents)

Parameters

output Tensor<T>

The controller output.

readContents List<Tensor<T>>

The current read contents.

Returns

Tensor<T>

The final output.

GenerateReadKeys(Tensor<T>)

Generates read keys for all read heads.

public List<Tensor<T>> GenerateReadKeys(Tensor<T> output)

Parameters

output Tensor<T>

The controller output.

Returns

List<Tensor<T>>

List of read keys.

GenerateWriteKey(Tensor<T>)

Generates write key.

public Tensor<T> GenerateWriteKey(Tensor<T> output)

Parameters

output Tensor<T>

The controller output.

Returns

Tensor<T>

The write key.

GetParameters()

Gets controller parameters.

public Vector<T> GetParameters()

Returns

Vector<T>

The parameter vector.

Reset()

Resets controller state.

public void Reset()

SetParameters(Vector<T>)

Sets controller parameters (updates internal weights).

public void SetParameters(Vector<T> parameters)

Parameters

parameters Vector<T>

The parameter vector to set.