Network pruning (1)

Title of the paper: Learning Efficient Convolutional Networks through Network Slimming
Topic of this paper: Network Slimming

(1) Three meanings of pruning:

1. reduce the model size;
2. decrease the run-time memory footprint;
3. lower the number of computing operations, without compromising accuracy.

1) Reduce model size; 2) Reduce runtime memory footprint; 3) Reduce the number of computational operations without compromising accuracy.

(2) Look directly at the conclusion:

It directly imposes sparsity-induced regularization on the scaling
factors in batch normalization layers, and unimportant channels can
thus be automatically identified during training and then pruned

Action object: BN layer (and unimportant channels)
Action mode: imposes sparsity-induced regularization on the scaling factorsp>

(3) What are Network Slimming and sparsity-induced regularization, scaling factors?????

First appearance of Network Slimming and sparsity-induced regularization, in this paragraph of the article.

The definition of Network Slimming is translated like this:

Our method applies L1 regularization to the scale factor in the batch normalization layer (BN), so it is easy to implement without introducing any changes to existing CNN architectures.Pushing the value of the BN scaling factor towards zero via L1 regularization allows us to identify unimportant channels (or neu-rons), since each scaling factor corresponds to a specific evolutionary channel (or neu-ron in a fully connected layer)yuan)

1) from the paragraph above (behind the blue line), 2) from the conclusion.
In other words:

1) Our approach imposes L1 regularlization on the scaling
factors in batch normalization (BN) layers
2) imposes sparsity-inducedregularization on the scaling factors

Then after reading this sentence, it is actually a picture

Then enter the Related Work paragraph
It mainly introduces the method of network structure compression, basically there is no need to watch hahahaha

Network slimming This paragraph
The papers sent by the Chinese understand the Chinese best!!!The bold type makes the point clear

I don't know if the translation is correct.Basically a penalty item