SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.06958 · 2020

Understanding the Role of Training Regimes in Continual Learning

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 20 functions out of this paper's own repositories and ran 8 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
imirzadeh/stable-continual-learning canonical 8 of 20
FunctionStatusWhere it lives
ResNet18 Ran imirzadeh/stable-continual-learning/stable_sgd/models.py
code served (permissive licence) · get_code("50bc0f11bc526034")
clone_variable_list Ran imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/utils.py
code served (permissive licence) · get_code("687d5d0cc60d8e5a")
conv3x3 Ran imirzadeh/stable-continual-learning/stable_sgd/models.py
code served (permissive licence) · get_code("583f9780bdd00a45")
create_conv_layer Ran imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/utils.py
code served (permissive licence) · get_code("4531aa012b415174")
create_fc_layer Ran imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/utils.py
code served (permissive licence) · get_code("f5ffbe87b7ec7cb6")
log_metrics Ran imirzadeh/stable-continual-learning/stable_sgd/utils.py
code served (permissive licence) · get_code("5d5a3e18f8eedade")
test_task_sequence Ran imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/fc_mnist.py
code served (permissive licence) · get_code("f6363cf3d882696c")
update_reservior Ran imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/er_utils.py
code served (permissive licence) · get_code("e401eb8ff332ab22")
bias_variable Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/model/model.py
code served (permissive licence) · get_code("a96da1b71ad1a9ef")
er_mem_update_hindsight Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/er_utils.py
code served (permissive licence) · get_code("c9a434336aceb23b")
get_permuted_mnist Not yet run imirzadeh/stable-continual-learning/stable_sgd/data_utils.py
code served (permissive licence) · get_code("dbc9c0571f3244fc")
get_permuted_mnist_tasks Not yet run imirzadeh/stable-continual-learning/stable_sgd/data_utils.py
code served (permissive licence) · get_code("193bb84a5b458862")
get_rotated_mnist Not yet run imirzadeh/stable-continual-learning/stable_sgd/data_utils.py
code served (permissive licence) · get_code("f3f11a0226a1466d")
image_scaling Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/data_utils.py
code served (permissive licence) · get_code("e75a0fb0c18a11fe")
init_experiment Not yet run imirzadeh/stable-continual-learning/stable_sgd/utils.py
code served (permissive licence) · get_code("adff890682647c1e")
random_crop_and_pad_image Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/data_utils.py
code served (permissive licence) · get_code("5b6dddd0f1e4b887")
random_horizontal_flip Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/data_utils.py
code served (permissive licence) · get_code("876c106a35f12456")
vgg_conv_layer Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/vgg_utils.py
code served (permissive licence) · get_code("30e387d6f5d1ab03")
vgg_fc_layer Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/utils/vgg_utils.py
code served (permissive licence) · get_code("dceae2df5c5c47ec")
weight_variable Not yet run imirzadeh/stable-continual-learning/external_libs/continual_learning_algorithms/model/model.py
code served (permissive licence) · get_code("70fae4aed55bdadc")

Repositories linked to this paper

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

Abstract

Catastrophic forgetting affects the training of neural networks, limiting their ability to learn multiple tasks sequentially. From the perspective of the well established plasticity-stability dilemma, neural networks tend to be overly plastic, lacking the stability necessary to prevent the forgetting of previous knowledge, which means that as learning progresses, networks tend to forget previously seen tasks. This phenomenon coined in the continual learning literature, has attracted much attention lately, and several families of approaches have been proposed with different degrees of success. However, there has been limited prior work extensively analyzing the impact that different training regimes -- learning rate, batch size, regularization method-- can have on forgetting. In this work, we depart from the typical approach of altering the learning algorithm to improve stability. Instead, we hypothesize that the geometrical properties of the local minima found for each task play an important role in the overall degree of forgetting. In particular, we study the effect of dropout, learning rate decay, and batch size, on forming training regimes that widen the tasks' local minima and consequently, on helping it not to forget catastrophically. Our study provides practical insights to improve stability via simple yet effective techniques that outperform alternative baselines.

For agents

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

get_harvested_code_for_paper("2006.06958")
get_code_for_paper("2006.06958")
have("2006.06958")

Connect an agent — have() is free.