SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2306.12230 · NeurIPS · 2023

Fantastic Weights and How to Find Them: Where to Prune in Dynamic Sparse Training

Jacek Tabor, Decebal Mocanu, Aleksandra Nowak, Bram Grooten

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 20 functions out of this paper's own repositories and ran 5 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
alooow/fantastic_weights_paper canonical 5 of 12
TimDettmers/sparse_learning canonical 0 of 8
FunctionStatusWhere it lives
magnitude_redistribution Ran alooow/fantastic_weights_paper/ImageNet/funcs.py
code served (permissive licence) · get_code("7fbf3f0c2be8ea47")
momentum_redistribution Ran alooow/fantastic_weights_paper/ImageNet/funcs.py
code served (permissive licence) · get_code("85588244473b88f9")
nonzero_redistribution Ran alooow/fantastic_weights_paper/ImageNet/funcs.py
code served (permissive licence) · get_code("e67bbb994ab5bae5")
snip_forward_conv2d Ran alooow/fantastic_weights_paper/ImageNet/snip.py
code served (permissive licence) · get_code("12d59cf9ada8b41a")
snip_forward_linear Ran alooow/fantastic_weights_paper/ImageNet/snip.py
code served (permissive licence) · get_code("94eeaaa88bd57941")
SNIP Not yet run alooow/fantastic_weights_paper/ImageNet/snip.py
code served (permissive licence) · get_code("04768d98372daca0")
build_resnet Not yet run TimDettmers/sparse_learning/imagenet/tuned_resnet/resnet.py
code served (permissive licence) · get_code("51a6a5c2a049e900")
build_resnet Not yet run alooow/fantastic_weights_paper/ImageNet/resnet.py
code served (permissive licence) · get_code("f069023ee573fe51")
build_resnet Not yet run alooow/fantastic_weights_paper/models/imagenet_resnet.py
code served (permissive licence) · get_code("61bbac980274b60d")
diverging_colors Not yet run TimDettmers/sparse_learning/plot_graphs.py
code served (permissive licence) · get_code("20c4b599af53b4ed")
fast_collate Not yet run alooow/fantastic_weights_paper/imagenet_main.py
code served (permissive licence) · get_code("717c76ec787ddfe7")
get_optimizer Not yet run alooow/fantastic_weights_paper/imagenet_main.py
code served (permissive licence) · get_code("410b25b6d0e7a081")
get_train_loader Not yet run alooow/fantastic_weights_paper/imagenet_main.py
code served (permissive licence) · get_code("04b0a36942f99acf")
initializations Not yet run alooow/fantastic_weights_paper/models/initializers.py
code served (permissive licence) · get_code("39e82a7c002940de")
rainbow_colors Not yet run TimDettmers/sparse_learning/plot_graphs.py
code served (permissive licence) · get_code("b8a31ec73b7a8b3c")
sequential_colors Not yet run TimDettmers/sparse_learning/plot_graphs.py
code served (permissive licence) · get_code("4746465c14058f8a")
uniform_coverage Not yet run TimDettmers/sparse_learning/imagenet/baseline/parameterized_tensors.py
code served (permissive licence) · get_code("bb11b6868648aca9")
variance_redistribution Not yet run TimDettmers/sparse_learning/mnist_cifar/extensions.py
code served (permissive licence) · get_code("177842ec69eed409")
your_pruning Not yet run TimDettmers/sparse_learning/mnist_cifar/extensions.py
code served (permissive licence) · get_code("77b9016770359ac8")
your_redistribution Not yet run TimDettmers/sparse_learning/mnist_cifar/extensions.py
code served (permissive licence) · get_code("37199b84c6e58a0e")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Dynamic Sparse Training (DST) is a rapidly evolving area of research that seeks to optimize the sparse initialization of a neural network by adapting its topology during training. It has been shown that under specific conditions, DST is able to outperform dense models. The key components of this framework are the pruning and growing criteria, which are repeatedly applied during the training process to adjust the network's sparse connectivity. While the growing criterion's impact on DST performance is relatively well studied, the influence of the pruning criterion remains overlooked. To address this issue, we design and perform an extensive empirical analysis of various pruning criteria to better understand their impact on the dynamics of DST solutions. Surprisingly, we find that most of the studied methods yield similar results. The differences become more significant in the low-density regime, where the best performance is predominantly given by the simplest technique: magnitude-based pruning.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2306.12230")
get_code_for_paper("2306.12230")
have("2306.12230")

Connect an agent — have() is free.