forward-forward-learning - Forward-Forward Algorithm for Local Layer-Wise Learning
Implements Hinton's Forward-Forward algorithm for local layer-wise neural network training without backpropagation.
Tags
Updated: 2026-09-23Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Train layers without backpropagation
- Compute layer goodness functions
- Generate self-contrastive negative samples
- Predict labels via maximum goodness
Inputs
- Positive data samples
- Negative data samples
- Class labels
- Network layer dimensions
Outputs
- Layer loss values
- Hidden layer activations
- Predicted labels
- Updated model parameters
Requirements
- PyTorch environment
- JAX and Flax framework