{"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/incremental-learning-through-deep-adaptation","title":"Incremental Learning Through Deep Adaptation","arxiv_id":"1705.04228","date":"2017-05-11","proceeding":"ICLR 2018 1","authors":["Amir Rosenfeld","John K. Tsotsos"],"abstract":"Given an existing trained neural network, it is often desirable to learn new\ncapabilities without hindering performance of those already learned. Existing\napproaches either learn sub-optimal solutions, require joint training, or incur\na substantial increment in the number of parameters for each added domain,\ntypically as many as the original network. We propose a method called\n\\emph{Deep Adaptation Networks} (DAN) that constrains newly learned filters to\nbe linear combinations of existing ones. DANs precisely preserve performance on\nthe original domain, require a fraction (typically 13\\%, dependent on network\narchitecture) of the number of parameters compared to standard fine-tuning\nprocedures and converge in less cycles of training to a comparable or better\nlevel of performance. When coupled with standard network quantization\ntechniques, we further reduce the parameter cost to around 3\\% of the original\nwith negligible or no loss in accuracy. The learned architecture can be\ncontrolled to switch between various learned representations, enabling a single\nnetwork to solve a task from multiple different domains. We conduct extensive\nexperiments showing the effectiveness of our method on a range of image\nclassification tasks and explore different aspects of its behavior.","url_abs":"http://arxiv.org/abs/1705.04228v2","url_pdf":"http://arxiv.org/pdf/1705.04228v2.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":[],"tasks":[{"task_slug":"continual-learning","task_name":"Continual Learning"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"incremental-learning","task_name":"Incremental Learning"},{"task_slug":"quantization","task_name":"Quantization"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/continual-learning-on-visual-domain-decathlon","task":"Continual Learning","dataset":"visual domain decathlon (10 tasks)","model":"DAN","rank_in_archive_order":7,"of":14,"metrics":{"Avg. Accuracy":"77.01","decathlon discipline (Score)":"2851"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1705.04228","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}