PHEW: Paths with Higher Edge-Weights give ''winning tickets'' without training data
Sparse neural networks have generated substantial interest recently because they can be more efficient in learning and inference, without any significant drop in performance. The "lottery ticket hypothesis" has showed the existence of such sparse subnetworks at initialization. Given a fully-connected initialized architecture, our aim is to find such "winning ticket" networks, without any training data. We first show the advantages of forming input-output paths, over pruning individual connections, to avoid bottlenecks in gradient propagation. Then, we show that Paths with Higher Edge-Weights (PHEW) at initialization have higher loss gradient magnitude, resulting in more efficient training. Selecting such paths can be performed without any data. We empirically validate the effectiveness of the proposed approach against pruning-before-training methods on CIFAR10, CIFAR100 and Tiny-ImageNet for VGG-Net and ResNet. PHEW achieves significant improvements on the current state-of-the-art methods at 10%, 5% and 2% network density. We also evaluate the structural similarity relationship between PHEW networks and pruned networks constructed through Iterated Magnitude Pruning (IMP), concluding that the former belong in the family of winning tickets networks.
Code (0)
등록된 구현이 없습니다.
Methods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
PHEW: Constructing Sparse Networks that Learn Fast and Generalize Well without Training Data
Methods that sparsify a network at initialization are important in practice because they greatly improve the efficiency of both learning and inference. Our work is based on a recently proposed decomposition of the Neural…
Pharmacogenomics in the Age of GWAS, Omics Atlases, and PheWAS
The search for causative pharmacogenomic loci is being transformed by integrative omics pipelines, but their outputs have only begun being applied to test design. We assess the direction of the field in light of Biobanks…
An Extended Symbolic-Arithmetic Model for Teaching Double-Black Removal with Rotation in Red-Black Trees
Double-black (DB) nodes have no place in red-black (RB) trees. So when DB nodes are formed, they are immediately removed. The removal of DB nodes that cause rotation and recoloring of other connected nodes poses greater …
K-shot NAS: Learnable Weight-Sharing for NAS with K-shot Supernets
In one-shot weight sharing for NAS, the weights of each operation (at each layer) are supposed to be identical for all architectures (paths) in the supernet. However, this rules out the possibility of adjusting operation…
Random Walks on Hypergraphs with Edge-Dependent Vertex Weights
Hypergraphs are used in machine learning to model higher-order relationships in data. While spectral methods for graphs are well-established, spectral theory for hypergraphs remains an active area of research. In this pa…
BIG-bench Machine Learning