{"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/rotational-unit-of-memory","title":"Rotational Unit of Memory","arxiv_id":"1710.09537","date":"2017-10-26","proceeding":"ICLR 2018 1","authors":["Rumen Dangovski","Li Jing","Marin Soljacic"],"abstract":"The concepts of unitary evolution matrices and associative memory have\nboosted the field of Recurrent Neural Networks (RNN) to state-of-the-art\nperformance in a variety of sequential tasks. However, RNN still have a limited\ncapacity to manipulate long-term memory. To bypass this weakness the most\nsuccessful applications of RNN use external techniques such as attention\nmechanisms. In this paper we propose a novel RNN model that unifies the\nstate-of-the-art approaches: Rotational Unit of Memory (RUM). The core of RUM\nis its rotational operation, which is, naturally, a unitary matrix, providing\narchitectures with the power to learn long-term dependencies by overcoming the\nvanishing and exploding gradients problem. Moreover, the rotational unit also\nserves as associative memory. We evaluate our model on synthetic memorization,\nquestion answering and language modeling tasks. RUM learns the Copying Memory\ntask completely and improves the state-of-the-art result in the Recall task.\nRUM's performance in the bAbI Question Answering task is comparable to that of\nmodels with attention mechanism. We also improve the state-of-the-art result to\n1.189 bits-per-character (BPC) loss in the Character Level Penn Treebank (PTB)\ntask, which is to signify the applications of RUM to real-world sequential\ndata. The universality of our construction, at the core of RNN, establishes RUM\nas a promising approach to language modeling, speech recognition and machine\ntranslation.","url_abs":"http://arxiv.org/abs/1710.09537v1","url_pdf":"http://arxiv.org/pdf/1710.09537v1.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":"rotational-unit-of-memory","repo_url":"https://github.com/jingli9111/RUM","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"rotational-unit-of-memory","repo_url":"https://github.com/jingli9111/RUM-Tensorflow","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"},{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"memorization","task_name":"Memorization"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":"speech-recognition","task_name":"Speech Recognition"},{"task_slug":"speech-recognition-1","task_name":"speech-recognition"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/question-answering-on-babi","task":"Question Answering","dataset":"bAbi","model":"RUM","rank_in_archive_order":8,"of":14,"metrics":{"Accuracy (trained on 1k)":"73.2%"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}