{"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/an-empirical-comparison-of-syllabuses-for","title":"An Empirical Comparison of Syllabuses for Curriculum Learning","arxiv_id":"1809.10789","date":"2018-09-27","proceeding":null,"authors":["Mark Collier","Joeran Beel"],"abstract":"Syllabuses for curriculum learning have been developed on an ad-hoc, per task\nbasis and little is known about the relative performance of different\nsyllabuses. We identify a number of syllabuses used in the literature. We\ncompare the identified syllabuses based on their effect on the speed of\nlearning and generalization ability of a LSTM network on three sequential\nlearning tasks. We find that the choice of syllabus has limited effect on the\ngeneralization ability of a trained network. In terms of speed of learning our\nresults demonstrate that the best syllabus is task dependent but that a\nrecently proposed automated curriculum learning approach - Predictive Gain,\nperforms very competitively against all identified hand-crafted syllabuses. The\nbest performing hand-crafted syllabus which we term Look Back and Forward\ncombines a syllabus which steps through tasks in the order of their difficulty\nwith a uniform distribution over all tasks. Our experimental results provide an\nempirical basis for the choice of syllabus on a new problem that could benefit\nfrom curriculum learning. Additionally, insights derived from our results shed\nlight on how to successfully design new syllabuses.","url_abs":"http://arxiv.org/abs/1809.10789v2","url_pdf":"http://arxiv.org/pdf/1809.10789v2.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":"an-empirical-comparison-of-syllabuses-for","repo_url":"https://github.com/MarkPKCollier/CurriculumLearningFYP","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[],"methods":[{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"speed","method_name":"SPEED"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}