Gradient clipping is a technique that bounds the magnitude of gradients computed during backpropagation to prevent them from exceeding a specified threshold. It is human-made: a deliberate algorithmic intervention designed to stabilize training of neural networks and other differentiable models. It persists as code implementing norm-based or value-based clipping strategies (global norm clipping, per-parameter clipping, clipping by value). [formal: clipatio graduum | substrate: behavior | horizon: a moment | explicit: yes | epoch: 1.02]
Accepted ontology entry
gradient clipping
Gradient clipping is a technique that bounds the magnitude of gradients computed during backpropagation to prevent them from exceeding a specified threshold. It is human-made: a deliberate algorithmic intervention designed to stabilize tra…
Definition
Why it is in scope
Gradient clipping is a human-made regularization technique used in training neural networks. It limits the magnitude of gradients during backpropagation to prevent exploding gradients, ensuring stable learning dynamics.
Names and aliases
- gradient clippingen · CANONICAL
Relations from this entry
- cmrwhb9js006isoacsf207m10INSTANCE_OF →
Gradient clipping IS a specific kind of regularization technique: it bounds gradient magnitudes to prevent exploding gradients during training. A competent speaker would call it 'a regularization method.' Files against the nearest kind (regularization) per Law 9.
- cmsllb31s07h4nobp1h0i7lu3DEPENDS_ON →
Removal test: remove optimization algorithms and gradient clipping stops operating — it constrains gradients before the parameter update step. Without the optimization algorithm to integrate into, gradient clipping has no operational context.
Relations to this entry
No accepted relations in this direction.
Record identity
- Created
- Aug 9, 2026, 9:02 AM UTC
- Content hash
- 680052db806ddf074db6bdcab1a29084c041fe25ff89e5a11c33210e1bb7f492