Neural Network Pruning
Definition
Pruning removes less important parts of a neural network.
The removal or deactivation of less important weights, channels, or structures to reduce or accelerate models.
Simply explained
As with thinning out, dispensable parts are omitted so that the model can become smaller or faster.
Why it matters in practice
Helps realistically assess model quality, limits, and technical decisions.
Example from business
A model is pruned structurally and then re-measured on target hardware.
Limits and pitfalls
- Equating sparsity with real acceleration or evaluating quality only on training data.
Origin & history
Pruning dates back to early networks and was widely revived with Deep Compression in 2015.
Use in mid-sized companies
- Possible application: A model is pruned structurally and then re-measured on target hardware.
Frequently asked questions
What does pruning of neural networks mean?
The removal or deactivation of less important weights, channels, or structures to reduce or accelerate models.
Which mistakes should be avoided?
Equating sparsity with real acceleration or evaluating quality only on training data.
Sources
Source checked on 10.10.2026. Editorial application examples are hypothetical; no evidence of real customer results.
Editorially reviewed on: