Dies ist eine Übersichtsseite mit Metadaten zu dieser wissenschaftlichen Arbeit. Der vollständige Artikel ist beim Verlag verfügbar.
How many winning tickets are there in one DNN?
2
Zitationen
2
Autoren
2020
Jahr
Abstract
The recent lottery ticket hypothesis proposes that there is one sub-network that matches the accuracy of the original network when trained in isolation. We show that instead each network contains several winning tickets, even if the initial weights are fixed. The resulting winning sub-networks are not instances of the same network under weight space symmetry, and show no overlap or correlation significantly larger than expected by chance. If randomness during training is decreased, overlaps higher than chance occur, even if the networks are trained on different tasks. We conclude that there is rather a distribution over capable sub-networks, as opposed to a single winning ticket.
Ähnliche Arbeiten
Deep Residual Learning for Image Recognition
2016 · 216.663 Zit.
U-Net: Convolutional Networks for Biomedical Image Segmentation
2015 · 86.244 Zit.
ImageNet classification with deep convolutional neural networks
2017 · 75.550 Zit.
Very Deep Convolutional Networks for Large-Scale Image Recognition
2014 · 75.406 Zit.
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
2016 · 52.833 Zit.