3 ms·
I believe several papers have examined efficiently pruning neural networks, but neural death would be better branding ;) (https://arxiv.org/pdf/1506.02626v3.pdf
by cing 10y ago
I believe several papers have examined efficiently pruning neural networks, but neural death would be better branding ;) (https://arxiv.org/pdf/1506.02626v3.pdf https://arxiv.org/pdf/1506.02626v3.pdf https://arxiv.org/pdf/1510.00149v5.pdf https://arxiv.org/pdf/1510.00149v5.pdf)
- simonster 10y agoYann LeCun called a technique for pruning weights "optimal brain damage" back in 1990: http://yann.lecun.com/exdb/publis/pdf/lecun-90b.pdf http://yann.lecun.com/exdb/publis/pdf/lecun-90b.pdf
- SimonKinds 10y agoThere is also this one https://arxiv.org/abs/1506.02515 https://arxiv.org/abs/1506.02515 which takes pruning a step further to reduce the sparsity. Also this one https://arxiv.org/abs/1608.04493 https://arxiv.org/abs/1608.04493 which makes sure not to kill any neurons which proves to be useful at a later stage in the pruning process.