SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1612.00796 · Proceedings of the National Academy of Sciences of the United States of America · 2016

Overcoming catastrophic forgetting in neural networks

Demis Hassabis, Dharshan Kumaran, Raia Hadsell, Joel Veness, Razvan Pascanu, Agnieszka Grabska-Barwinska, James Kirkpatrick, Neil Rabinowitz, Claudia Clopath, Guillaume Desjardins, Kieran Milan, John Quan, and 2 more

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 22 functions out of this paper's own repositories and ran 14 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.

FunctionStatusWhere it lives
EWC Ran srvCodes/continual-learning-benchmark/train/ewc.py
code served (permissive licence) · get_code("3ca3b2ad5cf08e7a")
EWC Ran codelion/adaptive-classifier/src/adaptive_classifier/ewc.py
code served (permissive licence) · get_code("a1465fe6cdcb57d7")
EWCMultitask Ran kamsyn95/CL_DNN/models/ewc_modules.py
pointer only (licence: NONE) · get_code("6aad701239ed7577")
ExemplarsDataset Ran mmasana/FACIL/src/approach/ewc.py
code served (permissive licence) · get_code("39b8751422fdac64")
ExperimentLogger Ran mmasana/FACIL/src/approach/ewc.py
code served (permissive licence) · get_code("1d10e03aa48307d3")
GradMultiply Ran wjmacro/continualmt/fairseq/modules/grad_multiply.py
code served (permissive licence) · get_code("b26c848baf0f673d")
Inc_Learning_Appr Ran mmasana/FACIL/src/approach/ewc.py
code served (permissive licence) · get_code("a47c1a346ec772ca")
MemoryDataset Ran mmasana/FACIL/src/approach/ewc.py
code served (permissive licence) · get_code("c1ecdf9c6eed882e")
cauchy_naive Ran catid/never_forget/models/s4.py
code served (permissive licence) · get_code("b7d9d94554e93b1b")
compute_omega_grads_norm Ran wannabeOG/MAS-PyTorch/utils/mas_utils.py
code served (permissive licence) · get_code("a1703c053471f8d6")
log_vandermonde_naive Ran catid/never_forget/models/s4.py
code served (permissive licence) · get_code("e25911109ba29783")
one_epoch_baseline Ran clam004/intro_continual_learning/contlearn/gettrainer.py
code served (permissive licence) · get_code("4473672cd588ecd4")
test Ran clam004/intro_continual_learning/contlearn/gettrainer.py
code served (permissive licence) · get_code("426df1649f33777b")
var2device Ran clam004/intro_continual_learning/contlearn/gettrainer.py
code served (permissive licence) · get_code("aff2576823d89d61")
Appr Not yet run mmasana/FACIL/src/approach/ewc.py
code served (permissive licence) · get_code("c1596b67340a6f97")
Approach Not yet run Minhchuyentoancbn/Continual-Learning/UCL/approaches/ewc.py
pointer only (licence: NONE) · get_code("118dad0481c8d9e3")
EWC Not yet run stijani/elastic-weight-consolidation-tf2/module.py
pointer only (licence: NONE) · get_code("5ebc601b0d506a2c")
ElasticWeightConsolidation Not yet run shivamsaboo17/overcoming-catastrophic-forgetting-in-neural-networks/elastic_weight_consolidation.py
pointer only (licence: NONE) · get_code("9d058983f0e0d999")
elastic_weight_consolidation_training Not yet run clam004/intro_continual_learning/contlearn/gettrainer.py
code served (permissive licence) · get_code("cbe2eeebdcb6a48a")
log_vandermonde_transpose_naive Not yet run catid/never_forget/models/s4.py
code served (permissive licence) · get_code("7cfca855faa8e8dc")
read_servers_from_hostfile Not yet run catid/never_forget/command_client_grid.py
code served (permissive licence) · get_code("5e7be12170048fbb")
register_mapping Not yet run xduan7/hat-cl/hat/modules/utils.py
code served (permissive licence) · get_code("ca1097ecd04eb4a3")

Repositories linked to this paper

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

Abstract

The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Neural networks are not, in general, capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks which they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on the MNIST hand written digit dataset and by learning several Atari 2600 games sequentially.

For agents

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

get_harvested_code_for_paper("1612.00796")
get_code_for_paper("1612.00796")
have("1612.00796")

Connect an agent — have() is free.