{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,18]],"date-time":"2026-08-18T01:55:55Z","timestamp":1787018155031,"version":"3.56.0"},"reference-count":47,"publisher":"MIT Press","issue":"8","content-domain":{"domain":["direct.mit.edu"],"crossmark-restriction":true},"short-container-title":[],"published-print":{"date-parts":[[2025,7,17]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Reservoir computing information processing based on untrained recurrent neural networks with random connections is expected to depend on the nonlinear properties of the neurons and the resulting oscillatory, chaotic, or fixed-point dynamics of the network. However, the degree of nonlinearity required and the range of suitable dynamical regimes for a given task remain poorly understood. To clarify these issues, we study the classification accuracy of a reservoir computer in artificial tasks of varying complexity while tuning both the neuron\u2019s degree of nonlinearity and the reservoir\u2019s dynamical regime. We find that even with activation functions of extremely reduced nonlinearity, weak recurrent interactions, and small input signals, the reservoir can compute useful representations. These representations, detectable only in higher-order principal components, make complex classification tasks linearly separable for the readout layer. Increasing the recurrent coupling leads to spontaneous dynamical behavior. Nevertheless, some input-related computations can \u201cride on top\u201d of oscillatory or fixed-point attractors with little loss of accuracy, whereas chaotic dynamics often reduces task performance. By tuning the system through the full range of dynamical phases, we observe in several classification tasks that accuracy peaks at both the oscillatory\/chaotic and chaotic\/fixed-point phase boundaries, supporting the edge of chaos hypothesis. We also present a regression task with the opposite behavior. Our findings, particularly the robust weakly nonlinear operating regime, may offer new perspectives for both technical and biological neural networks with random connectivity.<\/jats:p>","DOI":"10.1162\/neco_a_01770","type":"journal-article","created":{"date-parts":[[2025,7,8]],"date-time":"2025-07-08T13:24:11Z","timestamp":1751981051000},"page":"1469-1504","update-policy":"https:\/\/doi.org\/10.1162\/mitpressjournals.corrections.policy","source":"Crossref","is-referenced-by-count":6,"title":["Nonlinear Neural Dynamics and Classification Accuracy in Reservoir Computing"],"prefix":"10.1162","volume":"37","author":[{"given":"Claus","family":"Metzner","sequence":"first","affiliation":[{"name":"Pattern-Recognition Lab, Friedrich-Alexander-University, 91058 Erlangen, Germany claus.metzner@fau.de"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Achim","family":"Schilling","sequence":"additional","affiliation":[{"name":"Pattern-Recognition Lab, Friedrich-Alexander-University, 91058 Erlangen, Germany achim.schilling@fau.de"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Andreas","family":"Maier","sequence":"additional","affiliation":[{"name":"Pattern-Recognition Lab, Friedrich-Alexander-University, 91058 Erlangen, Germany andreas.maier@fau.de"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Patrick","family":"Krauss","sequence":"additional","affiliation":[{"name":"Pattern-Recognition Lab, Friedrich-Alexander-University, 91058 Erlangen, Germany patrick.krauss@fau.de"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"281","published-online":{"date-parts":[[2025,7,17]]},"reference":[{"key":"2025072212154322100_bib1","first-page":"2320","article-title":"Universal approximation of flows of control systems by recurrent neural networks","volume-title":"Proceedings of the 62nd IEEE Conference on Decision and Control","author":"Aguiar","year":"2023"},{"issue":"1","key":"2025072212154322100_bib2","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1186\/s40537-021-00444-8","article-title":"Review of deep learning: Concepts, CNN architectures, challenges, applications, future directions","volume":"8","author":"Alzubaidi","year":"2021","journal-title":"Journal of Big Data"},{"key":"2025072212154322100_bib3","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1016\/j.conb.2017.06.003","article-title":"Recurrent neural networks as versatile tools of neuroscience research","volume":"46","author":"Barak","year":"2017","journal-title":"Current Opinion in Neurobiology"},{"key":"2025072212154322100_bib4","author":"B\u00f6nsel","year":"2021","journal-title":"Control of noise-induced coherent oscillations in time-delayed neural motifs"},{"issue":"5","key":"2025072212154322100_bib5","doi-asserted-by":"publisher","first-page":"749","DOI":"10.1038\/nn.4286","article-title":"Is cortical connectivity optimized for storing information?","volume":"19","author":"Brunel","year":"2016","journal-title":"Nature Neuroscience"},{"issue":"5","key":"2025072212154322100_bib6","doi-asserted-by":"crossref","first-page":"1272","DOI":"10.1162\/neco.2009.01-09-947","article-title":"Connectivity, dynamics, and memory in reservoir computing with binary and analog neurons","volume":"22","author":"B\u00fcsing","year":"2010","journal-title":"Neural Computation"},{"issue":"12","key":"2025072212154322100_bib7","doi-asserted-by":"crossref","DOI":"10.1063\/5.0038163","article-title":"Do reservoir computers work best at the edge of chaos?","volume":"30","author":"Carroll","year":"2020","journal-title":"Chaos: An Interdisciplinary Journal of Nonlinear Science"},{"issue":"112","key":"2025072212154322100_bib8","first-page":"1","article-title":"Edge of stability in echo state networks","volume":"24","author":"Ceni","year":"2023","journal-title":"Journal of Machine Learning Research"},{"issue":"1","key":"2025072212154322100_bib9","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1038\/srep00514","article-title":"Information processing capacity of dynamical systems","volume":"2","author":"Dambre","year":"2012","journal-title":"Scientific Reports"},{"key":"2025072212154322100_bib10","doi-asserted-by":"crossref","DOI":"10.1016\/j.neucom.2024.128679","article-title":"Comparison of reservoir computing topologies using the recurrent kernel approach","volume":"611","author":"D\u2019Inverno","year":"2025","journal-title":"Neurocomputing"},{"key":"2025072212154322100_bib11","doi-asserted-by":"publisher","DOI":"10.1016\/j.neunet.2024.106179","article-title":"Fading memory as inductive bias in residual recurrent networks","volume":"173","author":"Dubinin","year":"2024","journal-title":"Neural Networks"},{"issue":"4","key":"2025072212154322100_bib12","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2412830122","article-title":"The functional role of oscillatory dynamics in neocortical circuits: A computational perspective","volume":"122","author":"Effenberger","year":"2025","journal-title":"Proceedings of the National Academy of Sciences"},{"issue":"6","key":"2025072212154322100_bib13","doi-asserted-by":"publisher","first-page":"564","DOI":"10.1038\/s42256-022-00498-0","article-title":"Gradient-based learning drives robust representations in recurrent neural networks by balancing compression and expansion","volume":"4","author":"Farrell","year":"2022","journal-title":"Nature Machine Intelligence"},{"key":"2025072212154322100_bib14","doi-asserted-by":"publisher","first-page":"50","DOI":"10.1016\/j.neunet.2018.04.003","article-title":"Effect of dilution in asymmetric recurrent neural networks","volume":"104","author":"Folli","year":"2018","journal-title":"Neural Networks"},{"issue":"3","key":"2025072212154322100_bib15","first-page":"1234","article-title":"Sparsity in reservoir computing neural networks","volume":"34","author":"Gallicchio","year":"2023","journal-title":"IEEE Transactions on Neural Networks and Learning Systems"},{"key":"2025072212154322100_bib16","doi-asserted-by":"publisher","first-page":"305","DOI":"10.1016\/j.neunet.2020.05.007","article-title":"Sparsity through evolutionary pruning prevents neuronal networks from overfitting","volume":"128","author":"Gerum","year":"2020","journal-title":"Neural Networks"},{"key":"2025072212154322100_bib17","doi-asserted-by":"publisher","first-page":"10","DOI":"10.1016\/j.neunet.2021.01.025","article-title":"Fading memory echo state networks are universal","volume":"138","author":"Gonon","year":"2021","journal-title":"Neural Networks"},{"key":"2025072212154322100_bib18","first-page":"2663","article-title":"Understanding and controlling memory in recurrent neural networks","volume-title":"Proceedings of the International Conference on Machine Learning","author":"Haviv","year":"2019"},{"issue":"8","key":"2025072212154322100_bib19","doi-asserted-by":"publisher","first-page":"1735","DOI":"10.1162\/neco.1997.9.8.1735","article-title":"Long short-term memory","volume":"9","author":"Hochreiter","year":"1997","journal-title":"Neural Computation"},{"key":"2025072212154322100_bib20","doi-asserted-by":"publisher","first-page":"263","DOI":"10.1007\/s12559-015-9333-0","article-title":"What are extreme learning machines? Filling the gap between Frank Rosenblatt\u2019s dream and John von Neumann\u2019s puzzle","volume":"7","author":"Huang","year":"2015","journal-title":"Cognitive Computation"},{"key":"2025072212154322100_bib21","doi-asserted-by":"publisher","first-page":"29","DOI":"10.1016\/j.neucom.2016.12.111","article-title":"Noise-modulated neural networks as an application of stochastic resonance","volume":"277","author":"Ikemoto","year":"2018","journal-title":"Neurocomputing"},{"key":"2025072212154322100_bib22","author":"Jacobs","year":"2025","journal-title":"Traveling waves integrate spatial information into spectral representations"},{"key":"2025072212154322100_bib23","author":"Jaeger","year":"2001","journal-title":"The \u201cecho state\u201d approach to analysing and training recurrent neural networks\u2014with an erratum note"},{"key":"2025072212154322100_bib24","author":"Jaeger","year":"2014","journal-title":"Controlling recurrent neural networks by conceptors"},{"key":"2025072212154322100_bib25","author":"Keller","year":"2023","journal-title":"Traveling waves encode the recent past and enhance sequence learning"},{"key":"2025072212154322100_bib26","doi-asserted-by":"publisher","DOI":"10.3389\/fncom.2019.00064","article-title":"\u201cRecurrence resonance\u201d in three-neuron motifs","volume":"13","author":"Krauss","year":"2019","journal-title":"Frontiers in Computational Neuroscience"},{"issue":"4","key":"2025072212154322100_bib27","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0214541","article-title":"Weight statistics controls dynamics in recurrent neural networks","volume":"14","author":"Krauss","year":"2019","journal-title":"PLOS One"},{"key":"2025072212154322100_bib28","doi-asserted-by":"publisher","DOI":"10.3389\/fnins.2016.00597","article-title":"Stochastic resonance controlled upregulation of internal noise after hearing loss as a putative cause of tinnitus-related neuronal hyperactivity","volume":"10","author":"Krauss","year":"2016","journal-title":"Frontiers in Neuroscience"},{"key":"2025072212154322100_bib29","doi-asserted-by":"crossref","DOI":"10.3389\/fncom.2019.00005","article-title":"Analysis of structure and dynamics in three-neuron motifs","volume":"13","author":"Krauss","year":"2019","journal-title":"Frontiers in Computational Neuroscience"},{"issue":"1\u20133","key":"2025072212154322100_bib30","doi-asserted-by":"publisher","first-page":"12","DOI":"10.1016\/0167-2789(90)90064-V","article-title":"Computation at the edge of chaos: Phase transitions and emergent computation","volume":"42","author":"Langton","year":"1990","journal-title":"Physica D: Nonlinear Phenomena"},{"issue":"7553","key":"2025072212154322100_bib31","doi-asserted-by":"publisher","first-page":"436","DOI":"10.1038\/nature14539","article-title":"Deep learning","volume":"521","author":"LeCun","year":"2015","journal-title":"Nature"},{"issue":"11","key":"2025072212154322100_bib32","doi-asserted-by":"publisher","first-page":"2531","DOI":"10.1162\/089976602760407955","article-title":"Real-time computing without stable states: A new framework for neural computation based on perturbations","volume":"14","author":"Maass","year":"2002","journal-title":"Neural Computation"},{"key":"2025072212154322100_bib33","article-title":"Universality and individuality in neural dynamics across large populations of recurrent networks","volume-title":"Advances in neural information processing systems","author":"Maheswaranathan","year":"2019"},{"key":"2025072212154322100_bib34","doi-asserted-by":"publisher","DOI":"10.3389\/fncom.2022.876315","article-title":"Dynamics and information import in recurrent neural networks","volume":"16","author":"Metzner","year":"2022","journal-title":"Frontiers in Computational Neuroscience"},{"key":"2025072212154322100_bib35","doi-asserted-by":"publisher","DOI":"10.3389\/fcpxs.2024.1479417","article-title":"Recurrence resonance-noise-enhanced dynamics in recurrent neural networks","volume":"2","author":"Metzner","year":"2024","journal-title":"Frontiers in Complex Systems"},{"issue":"2","key":"2025072212154322100_bib36","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3605943","article-title":"Recent advances in natural language processing via large pre-trained language models: A survey","volume":"56","author":"Min","year":"2023","journal-title":"ACM Computing Surveys"},{"issue":"26","key":"2025072212154322100_bib37","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.69.3717","article-title":"Suppressing chaos in neural networks by noise","volume":"69","author":"Molgedey","year":"1992","journal-title":"Physical Review Letters"},{"key":"2025072212154322100_bib38","author":"Narang","year":"2017","journal-title":"Exploring sparsity in recurrent neural networks"},{"issue":"5\u20136","key":"2025072212154322100_bib39","doi-asserted-by":"crossref","first-page":"465","DOI":"10.1016\/S0928-4257(00)01092-5","article-title":"Dynamical coding of sensory information with competitive networks","volume":"94","author":"Rabinovich","year":"2000","journal-title":"Journal of Physiology\u2013Paris"},{"issue":"1","key":"2025072212154322100_bib40","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevE.82.011903","article-title":"Stimulus-dependent suppression of chaos in recurrent neural networks","volume":"82","author":"Rajan","year":"2010","journal-title":"Physical Review E"},{"key":"2025072212154322100_bib41","first-page":"632","article-title":"Recurrent neural networks are universal approximators","volume-title":"Proceedings of the International Conference on Artificial Neural Networks","author":"Sch\u00e4fer","year":"2006"},{"key":"2025072212154322100_bib42","doi-asserted-by":"publisher","DOI":"10.3389\/fnins.2022.908330","article-title":"Intrinsic noise improves speech recognition in a computational model of the auditory pathway","volume":"16","author":"Schilling","year":"2022","journal-title":"Frontiers in Neuroscience"},{"issue":"12","key":"2025072212154322100_bib43","doi-asserted-by":"publisher","first-page":"4809","DOI":"10.1093\/brain\/awad255","article-title":"Predictive coding and stochastic resonance as fundamental principles of auditory phantom perception","volume":"146","author":"Schilling","year":"2023","journal-title":"Brain"},{"key":"2025072212154322100_bib44","doi-asserted-by":"publisher","first-page":"139","DOI":"10.1016\/bs.pbr.2021.01.025","article-title":"The stochastic resonance model of auditory perception: A unified explanation of tinnitus development, Zwicker tone illusion, and residual inhibition","volume":"262","author":"Schilling","year":"2021","journal-title":"Progress in Brain Research"},{"issue":"4","key":"2025072212154322100_bib45","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevX.8.041029","article-title":"Optimal sequence memory in driven random networks","volume":"8","author":"Schuecker","year":"2018","journal-title":"Physical Review X"},{"issue":"3","key":"2025072212154322100_bib46","doi-asserted-by":"crossref","DOI":"10.1371\/journal.pbio.0030068","article-title":"Highly nonrandom features of synaptic connectivity in local cortical circuits","volume":"3","author":"Song","year":"2005","journal-title":"PLOS Biology"},{"issue":"6","key":"2025072212154322100_bib47","doi-asserted-by":"publisher","first-page":"1408","DOI":"10.1162\/NECO_a_00449","article-title":"Randomly connected networks have short temporal memory","volume":"25","author":"Wallace","year":"2013","journal-title":"Neural Computation"}],"container-title":["Neural Computation"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/direct.mit.edu\/neco\/article-pdf\/37\/8\/1469\/2531740\/neco_a_01770.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/direct.mit.edu\/neco\/article-pdf\/37\/8\/1469\/2531740\/neco_a_01770.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,7,22]],"date-time":"2025-07-22T16:15:49Z","timestamp":1753200949000},"score":1,"resource":{"primary":{"URL":"https:\/\/direct.mit.edu\/neco\/article\/37\/8\/1469\/131384\/Nonlinear-Neural-Dynamics-and-Classification"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,7,17]]},"references-count":47,"journal-issue":{"issue":"8","published-online":{"date-parts":[[2025,7,17]]},"published-print":{"date-parts":[[2025,7,17]]}},"URL":"https:\/\/doi.org\/10.1162\/neco_a_01770","relation":{},"ISSN":["0899-7667","1530-888X"],"issn-type":[{"value":"0899-7667","type":"print"},{"value":"1530-888X","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2025,8]]},"published":{"date-parts":[[2025,7,17]]}}}