kr.ac.kaist.ir.deep.layer

FullTensorLayer

class FullTensorLayer extends Rank3TensorLayer

Layer: Basic, Fully-connected Rank 3 Tensor Layer.

Note

v0 = a column vector
Q = Rank 3 Tensor with size out, in × in is its entry.
L = Rank 3 Tensor with size out, 1 × in is its entry.
b = out × 1 matrix.

output = f( v0'.Q.v0 + L.v0 + b )

Linear Supertypes
Rank3TensorLayer, Layer, Serializable, Serializable, (ScalarMatrix) ⇒ ScalarMatrix, AnyRef, Any
Ordering
  1. Alphabetic
  2. By inheritance
Inherited
  1. FullTensorLayer
  2. Rank3TensorLayer
  3. Layer
  4. Serializable
  5. Serializable
  6. Function1
  7. AnyRef
  8. Any
  1. Hide All
  2. Show all
Learn more about member selection
Visibility
  1. Public
  2. All

Instance Constructors

  1. new FullTensorLayer(IO: (Int, Int), act: Activation, quad: Seq[ScalarMatrix] = ..., lin: Seq[ScalarMatrix] = ..., const: ScalarMatrix = null)

    IO

    is a tuple of the number of input and output, i.e. (2 → 4)

    act

    is an activation function to be applied

    quad

    is initial quadratic-level weight matrix Q for the case that it is restored from JSON (default: Seq())

    lin

    is initial linear-level weight matrix L for the case that it is restored from JSON (default: Seq())

    const

    is initial bias weight matrix b for the case that it is restored from JSON (default: null)

Value Members

  1. def !(error: ScalarMatrix, input: ScalarMatrix, output: ScalarMatrix): ScalarMatrix

    Backward computation.

    Backward computation.

    error

    to be propagated ( dG / dF is propagated from higher layer )

    input

    of this layer (in this case, x = entry of dX / dw)

    output

    of this layer (in this case, y)

    returns

    propagated error (in this case, dG/dx )

    Attributes
    protected[kr.ac.kaist.ir.deep]
    Definition Classes
    Rank3TensorLayer → Layer
    Note

    Let this layer have function F composed with function X(x) = x1'.Q.x2 + L.x + b and higher layer have function G. (Each output is treated as separately except propagation)

    Weight is updated with: dG/dW and propagate dG/dx

    For the computation, we only used denominator layout. (cf. Wikipedia Page of Matrix Computation) For the computation rules, see "Matrix Cookbook" from MIT.

  2. final def !=(arg0: AnyRef): Boolean

    Definition Classes
    AnyRef

  3. final def !=(arg0: Any): Boolean

    Definition Classes
    Any

  4. final def ##(): Int

    Definition Classes
    AnyRef → Any

  5. final def ==(arg0: AnyRef): Boolean

    Definition Classes
    AnyRef

  6. final def ==(arg0: Any): Boolean

    Definition Classes
    Any

  7. def >>:(x: ScalarMatrix): ScalarMatrix

    Sugar: Forward computation.

    Sugar: Forward computation. Calls apply(x)

    x

    input matrix

    returns

    output matrix

    Attributes
    protected[kr.ac.kaist.ir.deep]
    Definition Classes
    Layer

  8. def W: Seq[ScalarMatrix]

    weights for update

    weights for update

    returns

    weights

    Definition Classes
    Rank3TensorLayer → Layer

  9. val act: Activation

    is an activation function to be applied

    is an activation function to be applied

    Attributes
    protected
    Definition Classes
    FullTensorLayer → Rank3TensorLayer → Layer

  10. def andThen[A](g: (ScalarMatrix) ⇒ A): (ScalarMatrix) ⇒ A

    Definition Classes
    Function1
    Annotations
    @unspecialized()

  11. def apply(x: ScalarMatrix): ScalarMatrix

    Forward computation

    Forward computation

    x

    input matrix

    returns

    output matrix

    Definition Classes
    Rank3TensorLayer → Layer → Function1

  12. final def asInstanceOf[T0]: T0

    Definition Classes
    Any

  13. val bias: ScalarMatrix

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  14. def clone(): AnyRef

    Attributes
    protected[java.lang]
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

  15. def compose[A](g: (A) ⇒ ScalarMatrix): (A) ⇒ ScalarMatrix

    Definition Classes
    Function1
    Annotations
    @unspecialized()

  16. val dL: Seq[ScalarMatrix]

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  17. val dQ: Seq[ScalarMatrix]

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  18. def dW: Seq[ScalarMatrix]

    accumulated delta values

    accumulated delta values

    returns

    delta-weight

    Definition Classes
    Rank3TensorLayer → Layer

  19. val db: ScalarMatrix

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  20. final def eq(arg0: AnyRef): Boolean

    Definition Classes
    AnyRef

  21. def equals(arg0: Any): Boolean

    Definition Classes
    AnyRef → Any

  22. val fanIn: Int

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  23. val fanInA: Int

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  24. val fanInB: Int

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  25. val fanIns: (Int, Int, Int)

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  26. val fanOut: Int

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  27. def finalize(): Unit

    Attributes
    protected[java.lang]
    Definition Classes
    AnyRef
    Annotations
    @throws( classOf[java.lang.Throwable] )

  28. final def getClass(): Class[_]

    Definition Classes
    AnyRef → Any

  29. def hashCode(): Int

    Definition Classes
    AnyRef → Any

  30. def in1(x: ScalarMatrix): ScalarMatrix

    Retrieve first input

    Retrieve first input

    x

    input to be separated

    returns

    first input

    Attributes
    protected
    Definition Classes
    FullTensorLayer → Rank3TensorLayer

  31. def in2(x: ScalarMatrix): ScalarMatrix

    Retrive second input

    Retrive second input

    x

    input to be separated

    returns

    second input

    Attributes
    protected
    Definition Classes
    FullTensorLayer → Rank3TensorLayer

  32. final def isInstanceOf[T0]: Boolean

    Definition Classes
    Any

  33. val linear: Seq[ScalarMatrix]

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  34. final def ne(arg0: AnyRef): Boolean

    Definition Classes
    AnyRef

  35. final def notify(): Unit

    Definition Classes
    AnyRef

  36. final def notifyAll(): Unit

    Definition Classes
    AnyRef

  37. val quadratic: Seq[ScalarMatrix]

    Attributes
    protected
    Definition Classes
    Rank3TensorLayer

  38. final def synchronized[T0](arg0: ⇒ T0): T0

    Definition Classes
    AnyRef

  39. def toJSON: JsObject

    Translate this layer into JSON object (in Play! framework)

    Translate this layer into JSON object (in Play! framework)

    returns

    JSON object describes this layer

    Definition Classes
    FullTensorLayer → Layer

  40. def toString(): String

    Definition Classes
    Function1 → AnyRef → Any

  41. final def wait(): Unit

    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

  42. final def wait(arg0: Long, arg1: Int): Unit

    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

  43. final def wait(arg0: Long): Unit

    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

Inherited from Rank3TensorLayer

Inherited from Layer

Inherited from Serializable

Inherited from Serializable

Inherited from (ScalarMatrix) ⇒ ScalarMatrix

Inherited from AnyRef

Inherited from Any

Ungrouped