AutoDiff
The AutoDiff class supports Automatic Differentiation.
Attributes
- Graph
-
- Supertypes
-
class Objecttrait Matchableclass Any
Members list
Value members
Concrete methods
Add node to the computation graph.
Add node to the computation graph.
Value parameters
- node
-
the node to add to the pipeline
Attributes
Make a backward pass through the computation graph, computing partial derivatives and adjoints using reverse mode automatic differentiation. Compute adjoint: v-bar-sub-i = vb_i = vb_i+1 * d_v_i+1 / d_v_i
Make a backward pass through the computation graph, computing partial derivatives and adjoints using reverse mode automatic differentiation. Compute adjoint: v-bar-sub-i = vb_i = vb_i+1 * d_v_i+1 / d_v_i
Value parameters
- e
-
the negative error matrix
Attributes
- See also
-
https://www.jmlr.org/papers/volume18/17-468/17-468.pdf, section 3.2 Reverse Mode
Make a forward pass through the computation graph, computing output values based on the calculation v = f(u + b(0)) where u is the weighted input, b(0) is the bias vector and f is the activation function.
Make a forward pass through the computation graph, computing output values based on the calculation v = f(u + b(0)) where u is the weighted input, b(0) is the bias vector and f is the activation function.