{"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/asynchronous-training-of-word-embeddings-for","title":"Asynchronous Training of Word Embeddings for Large Text Corpora","arxiv_id":"1812.03825","date":"2018-12-07","proceeding":null,"authors":["Avishek Anand","Megha Khosla","Jaspreet Singh","Jan-Hendrik Zab","Zijian Zhang"],"abstract":"Word embeddings are a powerful approach for analyzing language and have been\nwidely popular in numerous tasks in information retrieval and text mining.\nTraining embeddings over huge corpora is computationally expensive because the\ninput is typically sequentially processed and parameters are synchronously\nupdated. Distributed architectures for asynchronous training that have been\nproposed either focus on scaling vocabulary sizes and dimensionality or suffer\nfrom expensive synchronization latencies.\n  In this paper, we propose a scalable approach to train word embeddings by\npartitioning the input space instead in order to scale to massive text corpora\nwhile not sacrificing the performance of the embeddings. Our training procedure\ndoes not involve any parameter synchronization except a final sub-model merge\nphase that typically executes in a few minutes. Our distributed training scales\nseamlessly to large corpus sizes and we get comparable and sometimes even up to\n45% performance improvement in a variety of NLP benchmarks using models trained\nby our distributed procedure which requires $1/10$ of the time taken by the\nbaseline approach. Finally we also show that we are robust to missing words in\nsub-models and are able to effectively reconstruct word representations.","url_abs":"http://arxiv.org/abs/1812.03825v1","url_pdf":"http://arxiv.org/pdf/1812.03825v1.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":"asynchronous-training-of-word-embeddings-for","repo_url":"https://github.com/jhzab/dist_w2v","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"information-retrieval","task_name":"Information Retrieval"},{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":"word-embeddings","task_name":"Word Embeddings"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}