{"id":13738632,"url":"https://github.com/imirzadeh/stable-continual-learning","last_synced_at":"2025-05-08T16:35:02.126Z","repository":{"id":37650973,"uuid":"257161410","full_name":"imirzadeh/stable-continual-learning","owner":"imirzadeh","description":"Towards increasing stability of neural networks for continual learning: https://arxiv.org/abs/2006.06958.pdf (NeurIPS'20)","archived":true,"fork":false,"pushed_at":"2023-03-25T00:01:57.000Z","size":528,"stargazers_count":75,"open_issues_count":6,"forks_count":11,"subscribers_count":6,"default_branch":"master","last_synced_at":"2024-11-15T07:34:58.042Z","etag":null,"topics":["catastrophic-forgetting","continual-learning","deep-learning","lifelong-learning","pytorch"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/imirzadeh.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null}},"created_at":"2020-04-20T03:28:25.000Z","updated_at":"2024-11-13T06:03:27.000Z","dependencies_parsed_at":"2024-01-15T04:12:41.092Z","dependency_job_id":null,"html_url":"https://github.com/imirzadeh/stable-continual-learning","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/imirzadeh%2Fstable-continual-learning","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/imirzadeh%2Fstable-continual-learning/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/imirzadeh%2Fstable-continual-learning/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/imirzadeh%2Fstable-continual-learning/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/imirzadeh","download_url":"https://codeload.github.com/imirzadeh/stable-continual-learning/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":253105712,"owners_count":21855084,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["catastrophic-forgetting","continual-learning","deep-learning","lifelong-learning","pytorch"],"created_at":"2024-08-03T03:02:29.858Z","updated_at":"2025-05-08T16:35:01.750Z","avatar_url":"https://github.com/imirzadeh.png","language":"Python","funding_links":[],"categories":["Python"],"sub_categories":[],"readme":"# Understanding the Role of Training Regimes in Continual Learning\nTowards increasing stability of neural networks for continual learning (NeurIPS'20)\n\n**Note: I will add an updated version of the code soon. If you have problem reproducing the results, please see the instructions for reproducing [experiment 1](https://github.com/imirzadeh/stable-continual-learning/issues/1) and [experiment 2](https://github.com/imirzadeh/stable-continual-learning/issues/5).**\n\n\n## 1. Code Structure\nThe high level structure of the code is as follows:\n\n```\nroot\n├── stable_sgd\n│   \n└── external_libs\n    └── continual_learning_algorithms\n    └── hessian_eigenthings\n```\n\n1. `stable_sgd`   : implementations of our stable and plastic training regimen for SGD (in Pytorch).      \n2. `external_libs`: third-party implementations we used for our experiments such as:   \n    2.1 `continual_learning_algorithms` Open source implementations for A-GEM, ER-Reservoir, and EWC (in Tensorflow).   \n    2.2 `hessian_eigenthings`: Open source implementation of deflated power iteration for eigenspectrum calculations (in Pytorch).  \n\n## 2. Setup \u0026 Installation\nThe code is tested on Python 3.6+, PyTorch 1.5.0, and Tensorflow 1.15.2. In addition, there are some other numerical and visualization libraries that are included in ``requirements.txt`` file. However, for convenience, we provide a script for setup:   \n```\nbash setup_and_install.sh\n```\n\n## 3. Replicating the Results\nNote: I will add an updated version of the code soon. If you have problem reproducing the results, please see the instructions for reproducing [experiment 1](https://github.com/imirzadeh/stable-continual-learning/issues/1) and [experiment 2](https://github.com/imirzadeh/stable-continual-learning/issues/5).\n\nWe provide scripts to replicate the results:   \n * 3.1 Run ```bash replicate_experiment_1.sh``` for experiment 1 (stable vs plastic).   \n * 3.2 Run ```bash replicate_experiment_2.sh``` for experiment 2 (Comparison with other methods with 20 tasks).\n * 3.3 Run ```bash replicate_appendix_c5.sh```  for the experiment in appendix C5 (Stabilizing other methods).\n \nFor faster replication, here we have only 3 runs per method per experiment, but we used 5 runs for the reported results.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fimirzadeh%2Fstable-continual-learning","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fimirzadeh%2Fstable-continual-learning","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fimirzadeh%2Fstable-continual-learning/lists"}