SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1803.05407 · 2018

Averaging Weights Leads to Wider Optima and Better Generalization

Pavel Izmailov, Andrew Wilson, Dmitry Vetrov, Timur Garipov, Dmitrii Podoprikhin

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 9 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
timgaripov/swa canonical 2 of 2
dice-group/aswa pwc_unofficial 3 of 6
zlwangustc/SWA_paddle pwc_unofficial 0 of 1
FunctionStatusWhere it lives
adjust_learning_rate Ran dice-group/aswa/utils.py
code served (permissive licence) · get_code("14ebf34e2f003912")
check_bn Ran dice-group/aswa/utils.py
code served (permissive licence) · get_code("6e59364a7cf6008f")
conv3x3 Ran timgaripov/swa/models/wide_resnet.py
code served (permissive licence) · get_code("00e569acd6b45ef0")
conv3x3 Ran timgaripov/swa/models/preresnet.py
code served (permissive licence) · get_code("583f9780bdd00a45")
make_layers Ran dice-group/aswa/models/vgg.py
code served (permissive licence) · get_code("4a29d37fffcfd8b1")
adjust_learning_rate Not yet run zlwangustc/SWA_paddle/utils.py
code served (permissive licence) · get_code("8d543b4110270e15")
eval Not yet run dice-group/aswa/utils.py
code served (permissive licence) · get_code("ff618422d0032e4e")
schedule Not yet run dice-group/aswa/ddp_train.py
code served (permissive licence) · get_code("24584656ffcaea3e")
selected_epochs Not yet run dice-group/aswa/analysis.py
code served (permissive licence) · get_code("fd2f0d7779c8d75b")

Repositories linked to this paper

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

Abstract

Deep neural networks are typically trained by optimizing a loss function with an SGD variant, in conjunction with a decaying learning rate, until convergence. We show that simple averaging of multiple points along the trajectory of SGD, with a cyclical or constant learning rate, leads to better generalization than conventional training. We also show that this Stochastic Weight Averaging (SWA) procedure finds much flatter solutions than SGD, and approximates the recent Fast Geometric Ensembling (FGE) approach with a single model. Using SWA we achieve notable improvement in test accuracy over conventional SGD training on a range of state-of-the-art residual networks, PyramidNets, DenseNets, and Shake-Shake networks on CIFAR-10, CIFAR-100, and ImageNet. In short, SWA is extremely easy to implement, improves generalization, and has almost no computational overhead.

For agents

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

get_harvested_code_for_paper("1803.05407")
get_code_for_paper("1803.05407")
have("1803.05407")

Connect an agent — have() is free.