Skip to main content

Table 1 Unstructured and structured pruning methods

From: A statistical approach for neural network pruning with application to internet of things

Methods

Advantages

Disadvantages

Unstructured [9,10,11]

Sparse structure original performance

High complexity and inefficiency on hardware

Structured [12,13,14,15, 22]

Prune weights in a predictable way

May not maintain unpruned performance