Papers of the day   All papers

Neural Networks are Surprisingly Modular

Comments

hardmaru: Neural Networks are Surprisingly Modular “Trained networks that undergo weight pruning are often significantly more modular than random networks with the same distribution of weights. Interestingly, they are much more modular when trained with dropout.“ https://arxiv.org/abs/2003.04881 https://t.co/uhLMewimAd

6 replies, 564 likes


Jeff Clune: Neural Networks are Surprisingly Modular. https://arxiv.org/abs/2003.04881

7 replies, 272 likes


Carlos E. Perez: New research that shows that artificial neural networks are surprisingly modular. Modularity is, in fact, a key principle in evolution. https://arxiv.org/abs/2003.04881 . .

0 replies, 31 likes


Bojan Tunguz: NNs seem surprisingly robust.

0 replies, 8 likes


Marco Salvi: Surprise!

0 replies, 5 likes


Bhagirath Kumar Lader: Neural Networks are Surprisingly Modular Daniel Filan, Shlomi Hod, Cody Wild, Andrew Critch, Stuart Russell #NeuralNetworks #modularity https://arxiv.org/pdf/2003.04881

0 replies, 1 likes


Content

Found on Mar 22 2020 at https://arxiv.org/pdf/2003.04881.pdf

PDF content of a computer science paper: Neural Networks are Surprisingly Modular