{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/generative-replay-with-feedback-connections","title":"Generative replay with feedback connections as a general strategy for continual learning","arxiv_id":"1809.10635","date":"2018-09-27","proceeding":null,"authors":["Gido M. van de Ven","Andreas S. Tolias"],"abstract":"A major obstacle to developing artificial intelligence applications capable\nof true lifelong learning is that artificial neural networks quickly or\ncatastrophically forget previously learned tasks when trained on a new one.\nNumerous methods for alleviating catastrophic forgetting are currently being\nproposed, but differences in evaluation protocols make it difficult to directly\ncompare their performance. To enable more meaningful comparisons, here we\nidentified three distinct scenarios for continual learning based on whether\ntask identity is known and, if it is not, whether it needs to be inferred.\nPerforming the split and permuted MNIST task protocols according to each of\nthese scenarios, we found that regularization-based approaches (e.g., elastic\nweight consolidation) failed when task identity needed to be inferred. In\ncontrast, generative replay combined with distillation (i.e., using class\nprobabilities as \"soft targets\") achieved superior performance in all three\nscenarios. Addressing the issue of efficiency, we reduced the computational\ncost of generative replay by integrating the generative model into the main\nmodel by equipping it with generative feedback or backward connections. This\nReplay-through-Feedback approach substantially shortened training time with no\nor negligible loss in performance. We believe this to be an important first\nstep towards making the powerful technique of generative replay scalable to\nreal-world continual learning applications.","url_abs":"http://arxiv.org/abs/1809.10635v2","url_pdf":"http://arxiv.org/pdf/1809.10635v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"generative-replay-with-feedback-connections","repo_url":"https://github.com/GMvandeVen/continual-learning","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"generative-replay-with-feedback-connections","repo_url":"https://github.com/WangTianduo/lifelong-learning-3-cases","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"generative-replay-with-feedback-connections","repo_url":"https://github.com/XSMUBC/DNC-lifelong-learning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"generative-replay-with-feedback-connections","repo_url":"https://github.com/XSMUBC/Lifelong-learning_xsm","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"generative-replay-with-feedback-connections","repo_url":"https://github.com/llzlcl/Continual-Learning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"continual-learning","task_name":"Continual Learning"},{"task_slug":"lifelong-learning","task_name":"Lifelong learning"},{"task_slug":null,"task_name":"Permuted-MNIST"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1809.10635","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}