{"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/exploring-embedding-methods-in-binary","title":"Exploring Embedding Methods in Binary Hyperdimensional Computing: A Case Study for Motor-Imagery based Brain-Computer Interfaces","arxiv_id":"1812.05705","date":"2018-12-13","proceeding":null,"authors":["Michael Hersche","José del R. Millán","Luca Benini","Abbas Rahimi"],"abstract":"Key properties of brain-inspired hyperdimensional (HD) computing make it a\nprime candidate for energy-efficient and fast learning in biosignal processing.\nThe main challenge is however to formulate embedding methods that map biosignal\nmeasures to a binary HD space. In this paper, we explore variety of such\nembedding methods and examine them with a challenging application of motor\nimagery brain-computer interface (MI-BCI) from electroencephalography (EEG)\nrecordings. We explore embedding methods including random projections,\nquantization based thermometer and Gray coding, and learning HD representations\nusing end-to-end training. All these methods, differing in complexity, aim to\nrepresent EEG signals in binary HD space, e.g. with 10,000 bits. This leads to\ndevelopment of a set of HD learning and classification methods that can be\nselectively chosen (or configured) based on accuracy and/or computational\ncomplexity requirements of a given task. We compare them with state-of-the-art\nlinear support vector machine (SVM) on an NVIDIA TX2 board using the 4-class\nBCI competition IV-2a dataset as well as a new 3-class dataset. Compared to\nSVM, results on 3-class dataset show that simple thermometer embedding achieves\nmoderate average accuracy (79.56% vs. 82.67%) with 26.8$\\times$ faster training\ntime and 22.3$\\times$ lower energy; on the other hand, switching to end-to-end\ntraining with learned HD representations wipes out these training benefits\nwhile boosting the accuracy to 84.22% (1.55% higher than SVM). Similar trend is\nobserved on the 4-class dataset where SVM achieves on average 74.29%: the\nthermometer embedding achieves 89.9$\\times$ faster training time and\n58.7$\\times$ lower energy, but a lower accuracy (67.09%) than the learned\nrepresentation of 72.54%.","url_abs":"http://arxiv.org/abs/1812.05705v2","url_pdf":"http://arxiv.org/pdf/1812.05705v2.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":"exploring-embedding-methods-in-binary","repo_url":"https://github.com/MHersche/HDembedding-BCI","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"brain-computer-interface","task_name":"Brain Computer Interface"},{"task_slug":"eeg-1","task_name":"EEG"},{"task_slug":"eeg","task_name":"Electroencephalogram (EEG)"},{"task_slug":"motor-imagery","task_name":"Motor Imagery"},{"task_slug":"quantization","task_name":"Quantization"}],"methods":[{"method_slug":"svm","method_name":"SVM"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}