Skip to main content

Multiply

Element-wise Multiplication

Multiplies two tensors element-by-element. Fundamental operation for gating mechanisms, attention weighting, and feature modulation.

Shape Contract:

  • Input a: [*shape] first tensor
  • Input b: [*shape] second tensor (must match shape of a)
  • Output: [*shape] element-wise product

Notes:

  • No learnable parameters (pure element-wise operation)
  • Both inputs must have identical shapes
  • Used in: gating (LSTM, GRU), attention weighting, feature scaling
  • Common pattern: x * sigmoid(gate) for learned gating
  • Supports broadcasting if shapes are compatible

Signature​

neuron Multiply()

Ports​

Inputs:

  • a: [*shape]
  • b: [*shape]

Outputs:

  • default: [*shape]

Implementation​

Source { source: "core", path: "structural/Multiply" }