Additional mathematical and learning concepts · MLD-1174
Straight-through estimator
A gradient approximation that substitutes a convenient backward derivative for a nondifferentiable or discrete forward operation.
Original reference page · detailed expansion pending. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.
Original example
Train through a quantization step.
Original caveat
The resulting gradient is generally biased.
Context and aliases
Level: Advanced. Stages: Theory · Training · Evaluation.
Tags: ML · DL.
Aliases: STE.
Related concepts
Concepts linking here
Original references
- Deep Learning: online textbook ↗ · Consulted
- Mathematics for Machine Learning ↗ · Reference link
- scikit-learn User Guide ↗ · Consulted
- Original reference URL ↗