{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,27]],"date-time":"2026-09-27T04:58:18Z","timestamp":1790485098367,"version":"4.1.0"},"reference-count":37,"publisher":"MDPI AG","issue":"19","license":[{"start":{"date-parts":[[2026,9,25]],"date-time":"2026-09-25T00:00:00Z","timestamp":1790294400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"name":"National Science and Technology Council","award":["NSTC-115-2221-E-324-016-"],"award-info":[{"award-number":["NSTC-115-2221-E-324-016-"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Sensors"],"abstract":"<jats:p>Deploying EEG classification networks on resource-constrained brain\u2013computer interface (BCI) hardware motivates structured filter pruning, yet which importance criterion to trust in already-compact architectures is unclear. We present a leakage-free, physically materialized comparison of five filter-importance criteria\u2014ablation-based structural sensitivity analysis (SSA), first-order Taylor, L1-norm, geometric median, and random\u2014across four datasets, two paradigms (motor imagery and ERP), two architectures, multiple capacities, and with and without fine-tuning. On EEGNet, after fine-tuning, SSA and Taylor are statistically equivalent: pooled over the four datasets their accuracy difference falls within a \u00b11 percentage-point margin (TOST, p=0.01), and on ShallowConvNet SSA is never worse than Taylor; both criteria significantly outperform the weight-based and random ones. SSA\u2019s value is therefore not a more accurate ranking but its formulation: on a weight-quantized (8-bit) EEGNet, where the loss gradient\u2014and hence Taylor\u2014is undefined on the quantized graph used here, SSA is the best gradient-free criterion, significantly beating L1, geometric-median, and random pruning in aggregate\u2014extending structured pruning to the quantized, gradient-free but editable deployment regime. We further show that the pruning schedule and fine-tuning matter more than the choice between strong criteria, and that training large then pruning with fine-tuning recovers accuracy equivalent to a directly trained smaller model. The results clarify where the importance criterion does and does not matter for compact EEG decoders.<\/jats:p>","DOI":"10.3390\/s26196085","type":"journal-article","created":{"date-parts":[[2026,9,25]],"date-time":"2026-09-25T10:34:10Z","timestamp":1790332450000},"page":"6085","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Gradient-Free Filter Importance for Structured Pruning of Compact EEG Networks"],"prefix":"10.3390","volume":"26","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-4197-2477","authenticated-orcid":false,"given":"Chu-Hui","family":"Lee","sequence":"first","affiliation":[{"name":"Department of Information Management, Chaoyang University of Technology, Taichung 413310, Taiwan"}],"role":[{"vocabulary":"crossref","role":"author"},{"role":"conceptualization","vocabulary":"credit"},{"role":"methodology","vocabulary":"credit"},{"role":"software","vocabulary":"credit"},{"role":"investigation","vocabulary":"credit"},{"role":"writing-original-draft","vocabulary":"credit"},{"role":"writing-review-editing","vocabulary":"credit"},{"role":"supervision","vocabulary":"credit"},{"role":"project-administration","vocabulary":"credit"},{"role":"funding-acquisition","vocabulary":"credit"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-7973-9112","authenticated-orcid":false,"given":"Chun-Ming","family":"Huang","sequence":"additional","affiliation":[{"name":"Taiwan Semiconductor Research Institute, National Institutes of Applied Research, Hsinchu 300091, Taiwan"}],"role":[{"vocabulary":"crossref","role":"author"},{"role":"conceptualization","vocabulary":"credit"},{"role":"methodology","vocabulary":"credit"},{"role":"software","vocabulary":"credit"},{"role":"validation","vocabulary":"credit"},{"role":"formal-analysis","vocabulary":"credit"},{"role":"resources","vocabulary":"credit"},{"role":"data-curation","vocabulary":"credit"},{"role":"writing-original-draft","vocabulary":"credit"},{"role":"writing-review-editing","vocabulary":"credit"},{"role":"visualization","vocabulary":"credit"}]}],"member":"1968","published-online":{"date-parts":[[2026,9,25]]},"reference":[{"key":"ref_1","doi-asserted-by":"crossref","first-page":"767","DOI":"10.1016\/S1388-2457(02)00057-3","article-title":"Brain\u2013computer interfaces for communication and control","volume":"113","author":"Wolpaw","year":"2002","journal-title":"Clin. Neurophysiol."},{"key":"ref_2","doi-asserted-by":"crossref","first-page":"031005","DOI":"10.1088\/1741-2552\/aab2f2","article-title":"A review of classification algorithms for EEG-based brain\u2013computer interfaces: A 10 year update","volume":"15","author":"Lotte","year":"2018","journal-title":"J. Neural Eng."},{"key":"ref_3","unstructured":"Ang, K.K., Chin, Z.Y., Zhang, H., and Guan, C. (2008). Filter bank common spatial pattern (FBCSP) in brain\u2013computer interface. IEEE International Joint Conference on Neural Networks (IJCNN), IEEE."},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"41","DOI":"10.1109\/MSP.2008.4408441","article-title":"Optimizing spatial filters for robust EEG single-trial analysis","volume":"25","author":"Blankertz","year":"2008","journal-title":"IEEE Signal Process. Mag."},{"key":"ref_5","doi-asserted-by":"crossref","first-page":"051001","DOI":"10.1088\/1741-2552\/ab260c","article-title":"Deep learning-based electroencephalography analysis: A systematic review","volume":"16","author":"Roy","year":"2019","journal-title":"J. Neural Eng."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"031001","DOI":"10.1088\/1741-2552\/ab0ab5","article-title":"Deep learning for electroencephalogram (EEG) classification tasks: A review","volume":"16","author":"Craik","year":"2019","journal-title":"J. Neural Eng."},{"key":"ref_7","doi-asserted-by":"crossref","first-page":"056013","DOI":"10.1088\/1741-2552\/aace8c","article-title":"EEGNet: A compact convolutional neural network for EEG-based brain\u2013computer interfaces","volume":"15","author":"Lawhern","year":"2018","journal-title":"J. Neural Eng."},{"key":"ref_8","doi-asserted-by":"crossref","first-page":"5391","DOI":"10.1002\/hbm.23730","article-title":"Deep learning with convolutional neural networks for EEG decoding and visualization","volume":"38","author":"Schirrmeister","year":"2017","journal-title":"Hum. Brain Mapp."},{"key":"ref_9","unstructured":"Han, S., Pool, J., Tran, J., and Dally, W.J. (2015, January 7\u201312). Learning both weights and connections for efficient neural networks. Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), Montreal, QC, Canada."},{"key":"ref_10","unstructured":"Li, H., Kadav, A., Durdanovic, I., Samet, H., and Graf, H.P. (2017, January 24\u201326). Pruning filters for efficient ConvNets. Proceedings of the International Conference on Learning Representations (ICLR), Toulon, France."},{"key":"ref_11","unstructured":"Molchanov, P., Tyree, S., Karras, T., Aila, T., and Kautz, J. (2017, January 24\u201326). Pruning convolutional neural networks for resource efficient inference. Proceedings of the International Conference on Learning Representations (ICLR), Toulon, France."},{"key":"ref_12","doi-asserted-by":"crossref","unstructured":"Molchanov, P., Mallya, A., Tyree, S., Frosio, I., and Kautz, J. (2019). Importance estimation for neural network pruning. IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE.","DOI":"10.1109\/CVPR.2019.01152"},{"key":"ref_13","doi-asserted-by":"crossref","unstructured":"He, Y., Liu, P., Wang, Z., Hu, Z., and Yang, Y. (2019). Filter pruning via geometric median for deep convolutional neural networks acceleration. IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE.","DOI":"10.1109\/CVPR.2019.00447"},{"key":"ref_14","unstructured":"Lee, C.H., and Huang, C.M. (2026, January 14\u201316). Ablation-based structural sensitivity analysis for EEG filter pruning. Proceedings of the International Conference on Knowledge Innovation and Invention (ICKII 2026), Engineering Proceedings, Sendai, Japan. in press."},{"key":"ref_15","doi-asserted-by":"crossref","first-page":"2773","DOI":"10.1109\/TNSRE.2020.3048106","article-title":"EEG-Inception: A novel deep convolutional neural network for assistive ERP-based brain\u2013computer interfaces","volume":"28","author":"Hornero","year":"2020","journal-title":"IEEE Trans. Neural Syst. Rehabil. Eng."},{"key":"ref_16","doi-asserted-by":"crossref","unstructured":"Zini, S., Bidone, F., and Napoletano, P. (2026). A robust multi-branch CNN-LSTM architecture for cross-subject motor imagery classification. Sensors, 26.","DOI":"10.3390\/s26113310"},{"key":"ref_17","doi-asserted-by":"crossref","unstructured":"Vafaei, E., and Hosseini, M. (2025). Transformers in EEG analysis: A review of architectures and applications in motor imagery, seizure, and emotion classification. Sensors, 25.","DOI":"10.3390\/s25051293"},{"key":"ref_18","doi-asserted-by":"crossref","unstructured":"Jacob, B., Kligys, S., Chen, B., Zhu, M., Tang, M., Howard, A., Adam, H., and Kalenichenko, D. (2018). Quantization and training of neural networks for efficient integer-arithmetic-only inference. IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE.","DOI":"10.1109\/CVPR.2018.00286"},{"key":"ref_19","unstructured":"Krishnamoorthi, R. (2018). Quantizing deep convolutional networks for efficient inference: A whitepaper. arXiv."},{"key":"ref_20","doi-asserted-by":"crossref","unstructured":"Wang, X., Hersche, M., T\u00f6mekce, B., Kaya, B., Magno, M., and Benini, L. (2020). An accurate EEGNet-based motor-imagery BCI for low-power edge computing. IEEE International Symposium on Medical Measurements and Applications (MeMeA), IEEE.","DOI":"10.1109\/MeMeA49120.2020.9137134"},{"key":"ref_21","unstructured":"Hinton, G., Vinyals, O., and Dean, J. (2015). Distilling the knowledge in a neural network. arXiv."},{"key":"ref_22","doi-asserted-by":"crossref","unstructured":"Shen, X., Zhong, H., Gu, Y., and Han, R. (2026). DO-PI-EATCNet: Efficient-attention- and dream-optimization-based channel selection for EEG motor imagery classification. Sensors, 26.","DOI":"10.3390\/s26113336"},{"key":"ref_23","unstructured":"LeCun, Y., Denker, J.S., and Solla, S.A. (1990, January 26\u201329). Optimal brain damage. Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), Denver, CO, USA."},{"key":"ref_24","unstructured":"Blalock, D., Gonzalez Ortiz, J.J., Frankle, J., and Guttag, J. (2020, January 2\u20134). What is the state of neural network pruning?. Proceedings of the Proceedings of Machine Learning and Systems (MLSys), Austin, TX, USA."},{"key":"ref_25","doi-asserted-by":"crossref","unstructured":"Yan, Y., Liang, S., and Xia, Y. (2026). OrdPrune-KD: An ordinal-consistency-based model compression framework for diabetic retinopathy grading. Sensors, 26.","DOI":"10.3390\/s26123636"},{"key":"ref_26","doi-asserted-by":"crossref","unstructured":"He, Y., Zhang, X., and Sun, J. (2017). Channel pruning for accelerating very deep neural networks. IEEE International Conference on Computer Vision (ICCV), IEEE.","DOI":"10.1109\/ICCV.2017.155"},{"key":"ref_27","doi-asserted-by":"crossref","unstructured":"Luo, J.H., Wu, J., and Lin, W. (2017). ThiNet: A filter level pruning method for deep neural network compression. IEEE International Conference on Computer Vision (ICCV), IEEE.","DOI":"10.1109\/ICCV.2017.541"},{"key":"ref_28","doi-asserted-by":"crossref","unstructured":"Lin, M., Ji, R., Wang, Y., Zhang, Y., Zhang, B., Tian, Y., and Shao, L. (2020). HRank: Filter pruning using high-rank feature map. IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE.","DOI":"10.1109\/CVPR42600.2020.00160"},{"key":"ref_29","unstructured":"Liu, Z., Sun, M., Zhou, T., Huang, G., and Darrell, T. (2019, January 6\u20139). Rethinking the value of network pruning. Proceedings of the International Conference on Learning Representations (ICLR), New Orleans, LA, USA."},{"key":"ref_30","unstructured":"Frankle, J., and Carbin, M. (2019, January 6\u20139). The lottery ticket hypothesis: Finding sparse, trainable neural networks. Proceedings of the International Conference on Learning Representations (ICLR), New Orleans, LA, USA."},{"key":"ref_31","doi-asserted-by":"crossref","unstructured":"Tangermann, M., M\u00fcller, K.R., Aertsen, A., Birbaumer, N., Braun, C., Brunner, C., Leeb, R., Mehring, C., Miller, K.J., and Mueller-Putz, G. (2012). Review of the BCI Competition IV. Front. Neurosci., 6.","DOI":"10.3389\/fnins.2012.00055"},{"key":"ref_32","unstructured":"Brunner, C., Leeb, R., M\u00fcller-Putz, G.R., Schl\u00f6gl, A., and Pfurtscheller, G. (2026, July 12). BCI Competition 2008\u2014Graz Data Set A (BCI IV-2a). BNCI Horizon 2020 Repository. Available online: http:\/\/bnci-horizon-2020.eu\/database\/data-sets\/001-2014\/."},{"key":"ref_33","unstructured":"Leeb, R., Brunner, C., M\u00fcller-Putz, G.R., Schl\u00f6gl, A., and Pfurtscheller, G. (2026, July 12). BCI Competition 2008\u2014Graz Data Set B (BCI IV-2b). BNCI Horizon 2020 Repository. Available online: http:\/\/bnci-horizon-2020.eu\/database\/data-sets\/004-2014\/."},{"key":"ref_34","doi-asserted-by":"crossref","first-page":"035008","DOI":"10.1088\/1741-2560\/11\/3\/035008","article-title":"Influence of P300 latency jitter on event related potential-based brain\u2013computer interface performance","volume":"11","author":"Aloise","year":"2014","journal-title":"J. Neural Eng."},{"key":"ref_35","doi-asserted-by":"crossref","first-page":"066011","DOI":"10.1088\/1741-2552\/aadea0","article-title":"MOABB: Trustworthy algorithm benchmarking for BCIs","volume":"15","author":"Jayaram","year":"2018","journal-title":"J. Neural Eng."},{"key":"ref_36","doi-asserted-by":"crossref","unstructured":"Gramfort, A., Luessi, M., Larson, E., Engemann, D.A., Strohmeier, D., Brodbeck, C., Goj, R., Jas, M., Brooks, T., and Parkkonen, L. (2013). MEG and EEG data analysis with MNE-Python. Front. Neurosci., 7.","DOI":"10.3389\/fnins.2013.00267"},{"key":"ref_37","unstructured":"Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., and Antiga, L. (2019, January 8\u201314). PyTorch: An imperative style, high-performance deep learning library. Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), Vancouver, BC, Canada."}],"container-title":["Sensors"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/1424-8220\/26\/19\/6085\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,27]],"date-time":"2026-09-27T04:31:54Z","timestamp":1790483514000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/1424-8220\/26\/19\/6085"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,25]]},"references-count":37,"journal-issue":{"issue":"19","published-online":{"date-parts":[[2026,10]]}},"alternative-id":["s26196085"],"URL":"https:\/\/doi.org\/10.3390\/s26196085","relation":{},"ISSN":["1424-8220"],"issn-type":[{"value":"1424-8220","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,25]]}}}