{"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/what-is-one-grain-of-sand-in-the-desert","title":"What Is One Grain of Sand in the Desert? Analyzing Individual Neurons in Deep NLP Models","arxiv_id":"1812.09355","date":"2018-12-21","proceeding":null,"authors":["Fahim Dalvi","Nadir Durrani","Hassan Sajjad","Yonatan Belinkov","Anthony Bau","James Glass"],"abstract":"Despite the remarkable evolution of deep neural networks in natural language\nprocessing (NLP), their interpretability remains a challenge. Previous work\nlargely focused on what these models learn at the representation level. We\nbreak this analysis down further and study individual dimensions (neurons) in\nthe vector representation learned by end-to-end neural models in NLP tasks. We\npropose two methods: Linguistic Correlation Analysis, based on a supervised\nmethod to extract the most relevant neurons with respect to an extrinsic task,\nand Cross-model Correlation Analysis, an unsupervised method to extract salient\nneurons w.r.t. the model itself. We evaluate the effectiveness of our\ntechniques by ablating the identified neurons and reevaluating the network's\nperformance for two tasks: neural machine translation (NMT) and neural language\nmodeling (NLM). We further present a comprehensive analysis of neurons with the\naim to address the following questions: i) how localized or distributed are\ndifferent linguistic properties in the models? ii) are certain neurons\nexclusive to some properties and not others? iii) is the information more or\nless distributed in NMT vs. NLM? and iv) how important are the neurons\nidentified through the linguistic correlation method to the overall task? Our\ncode is publicly available as part of the NeuroX toolkit (Dalvi et al. 2019).","url_abs":"http://arxiv.org/abs/1812.09355v1","url_pdf":"http://arxiv.org/pdf/1812.09355v1.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":"what-is-one-grain-of-sand-in-the-desert","repo_url":"https://github.com/fdalvi/NeuroX","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","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":"nmt","task_name":"NMT"},{"task_slug":"sand","task_name":"Sand"}],"methods":[{"method_slug":"interpretability","method_name":"Interpretability"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1812.09355","atlas_url":"https://app.syntology.ai/?focus=1812.09355","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}