Optimization and training dynamics · MLD-0480
Gradient explosion
Gradients growing to numerically or operationally destabilizing magnitudes.
Original reference page · detailed expansion pending. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.
Original example
Observe extreme updates in a recurrent network.
Original caveat
Clipping treats the update, not necessarily the root cause.
Context and aliases
Level: Intermediate. Stages: Training · Theory.
Tags: ML · DL · SLM · LLM.
Aliases: Exploding gradient.
Related concepts
Concepts linking here
Original references
- Deep Learning: online textbook ↗ · Consulted
- PyTorch optimization algorithms ↗ · Reference link
- PyTorch documentation ↗ · Reference link
- Original reference URL ↗