{"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/selfless-sequential-learning","title":"Selfless Sequential Learning","arxiv_id":"1806.05421","date":"2018-06-14","proceeding":"ICLR 2019 5","authors":["Rahaf Aljundi","Marcus Rohrbach","Tinne Tuytelaars"],"abstract":"Sequential learning, also called lifelong learning, studies the problem of\nlearning tasks in a sequence with access restricted to only the data of the\ncurrent task. In this paper we look at a scenario with fixed model capacity,\nand postulate that the learning process should not be selfish, i.e. it should\naccount for future tasks to be added and thus leave enough capacity for them.\nTo achieve Selfless Sequential Learning we study different regularization\nstrategies and activation functions. We find that imposing sparsity at the\nlevel of the representation (i.e.~neuron activations) is more beneficial for\nsequential learning than encouraging parameter sparsity. In particular, we\npropose a novel regularizer, that encourages representation sparsity by means\nof neural inhibition. It results in few active neurons which in turn leaves\nmore free neurons to be utilized by upcoming tasks. As neural inhibition over\nan entire layer can be too drastic, especially for complex tasks requiring\nstrong representations, our regularizer only inhibits other neurons in a local\nneighbourhood, inspired by lateral inhibition processes in the brain. We\ncombine our novel regularizer, with state-of-the-art lifelong learning methods\nthat penalize changes to important previously learned parts of the network. We\nshow that our new regularizer leads to increased sparsity which translates in\nconsistent performance improvement %over alternative regularizers we studied on\ndiverse datasets.","url_abs":"http://arxiv.org/abs/1806.05421v5","url_pdf":"http://arxiv.org/pdf/1806.05421v5.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":"selfless-sequential-learning","repo_url":"https://github.com/rahafaljundi/Selfless-Sequential-Learning","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"lifelong-learning","task_name":"Lifelong learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.05421","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}