{"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/deep-compressive-autoencoder-for-action","title":"Deep Compressive Autoencoder for Action Potential Compression in Large-Scale Neural Recording","arxiv_id":"1809.05522","date":"2018-09-14","proceeding":null,"authors":["Tong Wu","Wenfeng Zhao","Edward Keefer","Zhi Yang"],"abstract":"Understanding the coordinated activity underlying brain computations requires\nlarge-scale, simultaneous recordings from distributed neuronal structures at a\ncellular-level resolution. One major hurdle to design high-bandwidth,\nhigh-precision, large-scale neural interfaces lies in the formidable data\nstreams that are generated by the recorder chip and need to be online\ntransferred to a remote computer. The data rates can require hundreds to\nthousands of I/O pads on the recorder chip and power consumption on the order\nof Watts for data streaming alone. We developed a deep learning-based\ncompression model to reduce the data rate of multichannel action potentials.\nThe proposed model is built upon a deep compressive autoencoder (CAE) with\ndiscrete latent embeddings. The encoder is equipped with residual\ntransformations to extract representative features from spikes, which are\nmapped into the latent embedding space and updated via vector quantization\n(VQ). The decoder network reconstructs spike waveforms from the quantized\nlatent embeddings. Experimental results show that the proposed model\nconsistently outperforms conventional methods by achieving much higher\ncompression ratios (20-500x) and better or comparable reconstruction\naccuracies. Testing results also indicate that CAE is robust against a diverse\nrange of imperfections, such as waveform variation and spike misalignment, and\nhas minor influence on spike sorting accuracy. Furthermore, we have estimated\nthe hardware cost and real-time performance of CAE and shown that it could\nsupport thousands of recording channels simultaneously without excessive\npower/heat dissipation. The proposed model can reduce the required data\ntransmission bandwidth in large-scale recording experiments and maintain good\nsignal qualities. The code of this work has been made available at\nhttps://github.com/tong-wu-umn/spike-compression-autoencoder","url_abs":"http://arxiv.org/abs/1809.05522v2","url_pdf":"http://arxiv.org/pdf/1809.05522v2.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":"deep-compressive-autoencoder-for-action","repo_url":"https://github.com/tong-wu-umn/spike-compression-autoencoder","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"quantization","task_name":"Quantization"},{"task_slug":"spike-sorting","task_name":"Spike Sorting"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}