SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2010.04003 · 2020

A Theoretical Analysis of Catastrophic Forgetting through the NTK Overlap Matrix

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 13 functions out of this paper's own repositories and ran 7 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
tldoan/PCA-OGD canonical 7 of 13
FunctionStatusWhere it lives
LeNetC Ran tldoan/PCA-OGD/models/lenet.py
code served (permissive licence) · get_code("270673c93e064a59")
accuracy Ran tldoan/PCA-OGD/utils/utils.py
code served (permissive licence) · get_code("25decb913a86724b")
orthonormalize Ran tldoan/PCA-OGD/algos/ogd.py
code served (permissive licence) · get_code("5dd8f3694051c1e0")
penalty Ran tldoan/PCA-OGD/algos/ewc.py
code served (permissive licence) · get_code("62c3f89b98633488")
project2cone2 Ran tldoan/PCA-OGD/algos/gem.py
code served (permissive licence) · get_code("25f23f7771ae8223")
project_vec Ran tldoan/PCA-OGD/algos/ogd.py
code served (permissive licence) · get_code("1db8a5e0aca11e0d")
test_error Ran tldoan/PCA-OGD/utils/utils.py
code served (permissive licence) · get_code("6b253a0b38ed156a")
CIFAR10 Not yet run tldoan/PCA-OGD/dataloaders/base.py
code served (permissive licence) · get_code("b5b0b299677171a5")
CIFAR100 Not yet run tldoan/PCA-OGD/dataloaders/base.py
code served (permissive licence) · get_code("0d51ab16aa5bc90c")
MNIST Not yet run tldoan/PCA-OGD/dataloaders/base.py
code served (permissive licence) · get_code("72c9598b309e4e5a")
PermutedGen Not yet run tldoan/PCA-OGD/dataloaders/datasetGen.py
code served (permissive licence) · get_code("f9b8caa6279ce116")
RotatedGen Not yet run tldoan/PCA-OGD/dataloaders/datasetGen.py
code served (permissive licence) · get_code("25d61b37abe1981b")
SplitGen Not yet run tldoan/PCA-OGD/dataloaders/datasetGen.py
code served (permissive licence) · get_code("8c9829ef771c8fe8")

Repositories linked to this paper

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

Abstract

Continual learning (CL) is a setting in which an agent has to learn from an incoming stream of data during its entire lifetime. Although major advances have been made in the field, one recurring problem which remains unsolved is that of Catastrophic Forgetting (CF). While the issue has been extensively studied empirically, little attention has been paid from a theoretical angle. In this paper, we show that the impact of CF increases as two tasks increasingly align. We introduce a measure of task similarity called the NTK overlap matrix which is at the core of CF. We analyze common projected gradient algorithms and demonstrate how they mitigate forgetting. Then, we propose a variant of Orthogonal Gradient Descent (OGD) which leverages structure of the data through Principal Component Analysis (PCA). Experiments support our theoretical findings and show how our method can help reduce CF on classical CL datasets.

For agents

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

get_harvested_code_for_paper("2010.04003")
get_code_for_paper("2010.04003")
have("2010.04003")

Connect an agent — have() is free.