Class BackwardOp
- Namespace
- AiDotNet.JitCompiler.IR.Operations
- Assembly
- AiDotNet.dll
Base class for backward (gradient) operations in the IR.
public abstract class BackwardOp : IROp- Inheritance
-
BackwardOp
- Derived
- Inherited Members
Remarks
Backward operations compute gradients during backpropagation for training. Each forward operation has corresponding backward operation(s) that compute the gradient with respect to its inputs.
For Beginners: These operations compute gradients for training.
In neural network training:
- Forward pass: Compute outputs from inputs
- Backward pass: Compute how to adjust weights to reduce error
Backward operations implement the chain rule of calculus to flow gradients backward through the network.
Properties
SavedForwardTensorId
The tensor ID from the forward pass that may be needed for gradient computation. Many backward operations need the forward pass output or inputs.
public int? SavedForwardTensorId { get; set; }Property Value
- int?