Real-Time Computation at the Edge of Chaos in Recurrent Neural Networks
The DOI matches its Crossref record; 26 of the 26 references it cites were found in a registry or an index.
Abstract
Depending on the connectivity, recurrent networks of simple computational units can show very different types of dynamics, ranging from totally ordered to chaotic. We analyze how the type of dynamics (ordered or chaotic) exhibited by randomly connected networks of threshold gates driven by a time-varying input signal depends on the parameters describing the distribution of the connectivity matrix. In particular, we calculate the critical boundary in parameter space where the transition from ordered to chaotic dynamics takes place. Employing a recently developed framework for analyzing real-time computations, we show that only near the critical boundary can such networks perform complex computations on time series. Hence, this result strongly supports conjectures that dynamical systems that are capable of doing complex computational tasks should operate near the edge of chaos, that is, the transition from ordered to chaotic dynamics.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Intrinsic fluctuations that are poised ''at the edge of chaos'' are thought to maximize the computational efficiency of such systems and the flexibility to switch between adaptive regimes (Bertschinger & Natschla ¨ger, 2004).”
From A potential mechanism for Gibsonian resonance: behavioral entrainment emerges from local homeostasis in an unsupervised reservoir network (2024) · cites this paper as Bertschinger & Natschläger, 2004
In the citing paper
“Finally, we show that a randomly connected circuit model, akin to a “reservoir” network13-15, can reproduce the observed distribution of timescales, provided that the network operates at the critical point (or “edge of chaos”)16-18.”
From A reservoir of time constants for memory traces in cortical neurons (2011) · cites this paper as [17]
Citations
Cites
Self-organized criticality: An explanation of the 1/f noise
Bak, P., Tang, C., & Wiesenfeld, K. (1987) · Physical Review Letters · doi:10.1103/physrevlett.59.381 · Cited by 7,591
The origins of order; self organization and selection in evolution
(1994) · International Journal of Biochemistry · doi:10.1016/0020-711x(94)90119-8 · Cited by 6,649
Introduction to the theory of neural computation
(1994) · Neural Networks · doi:10.1016/0893-6080(94)90014-0 · Cited by 6,401
Bak, P., Tang, C., & Wiesenfeld, K. (1988) · Physical Review A · doi:10.1103/physreva.38.364 · Cited by 4,713
Maass, W., Natschläger, T., & Markram, H. (2002) · Neural Computation · doi:10.1162/089976602760407955 · Cited by 4,025 · cited 10×
“(Maass et al., 2002) and indepenently by Jaeger (Jaeger, 2002) who called it “echo state networks”.”
BackgroundInfluentialSelf-citationChaos in Neuronal Networks with Balanced Excitatory and Inhibitory Activity
van Vreeswijk, C., & Sompolinsky, H. (1996) · Science · doi:10.1126/science.274.5293.1724 · Cited by 1,811
Computation at the edge of chaos: Phase transitions and emergent computation
Langton, C. G. (1990) · Physica D: Nonlinear Phenomena · doi:10.1016/0167-2789(90)90064-v · Cited by 1,527 · cited 5×
“These results provide further evidence for the idea of “computation at the edge of chaos” (Langton, 1990).”
BackgroundResultInfluentialFading memory and the problem of approximating nonlinear operators with Volterra series
Boyd, S., & Chua, L. (1985) · IEEE Transactions on Circuits and Systems · doi:10.1109/tcs.1985.1085649 · Cited by 991
Random Networks of Automata: A Simple Annealed Approximation
Derrida, B., & Pomeau, Y. (1986) · Europhysics Letters (EPL) · doi:10.1209/0295-5075/1/2/001 · Cited by 706
Dynamic Patterns in Complex Systems
(1988) · doi:10.1142/0534 · Cited by 548
Research works printed but not linked: 1
- Mitchell M., 1993, Complex Systems, 7, 89
Cited by
Items 801–831 of 831
Delay asymmetry induces a functional separation between sensitivity and coordinationPreprint
Yamaki, N., & Churiki, T. (2026) · Research Square · doi:10.21203/rs.3.rs-8896814/v1
Current-control of chaos and effects of thermal fluctuations in magnetic tunnel junctions
Tatsumi, R., Miwa, S., Matsueda, H., & Chiba, T. (2026) · Physical Review Applied · doi:10.1103/nny3-5mnz
“As demonstrated in previous studies, it is often reported that reservoir computing achieves high computational performance when their elements exhibit “edge of chaos”, which is a transient state between the periodic and chaotic states [1, 10– 12].”
Structural and dynamical strategies to prevent runaway excitation in reservoir computing
Metzner, C., Schilling, A., Maier, A., Kinfe, T. M., & Krauß, P. (2026) · Frontiers in Computational Neuroscience · doi:10.3389/fncom.2026.1845838
“Early studies in reservoir computing and recurrent neural networks suggested that computational capability can peak near transitions between ordered and chaotic dynamical regimes (Bertschinger and Natschläger, 2004; Legenstein and Maass, 2007; Boedecker et al., 2012).”
McAllister, J., Houghton, C. J., Wade, J., & O’Donnell, C. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.03.30.715411 · Cited by 5
From dynamics to computations in recurrent neural networks
Mastrogiuseppe, F. (2017) · theses.fr (ABES) · doi:10.70675/8534b1a4z7c95z45efz9de8z52f51f68a178 · Cited by 3
From neural coding to decision-making
Berlemont, K. (2020) · theses.fr (ABES) · doi:10.70675/98ad90cazb967z4ef3zb51az8213de1b171f
Reservoir computing in lithium niobate on insulator platforms
Abdalla, M. (2024) · theses.fr (ABES) · doi:10.70675/c7771ae4z9ee6z4a4dz9ca1z2f428dbe4fa2
Neuro-inspired computing with excitable microlasers
Masominia, A. H. (2024) · theses.fr (ABES) · doi:10.70675/f448ec0bz8e07z4e59z8b05z1aef41e55b65 · Cited by 1
Memory uncertainty relation and harmonic memory in random recurrent networks
Haruna, T., & Nakajima, K. (2026) · Physical review. E · doi:10.1103/2vtp-m1yk
The architecture of agreement: from strings to structures
Krivochen, D. G., Saddy, D., & Franck, J. (2026) · Frontiers in Language Sciences · doi:10.3389/flang.2026.1689620
“Complexity and computation appear around unstable points, specifically in the transition between order and chaos, called the “edge of chaos” (Crutchfield and Young, 1989; Bertschinger and Natschläger, 2004).”
Song, K., Bonnin, M., Ascoli, A., & Corinto, F. (2026) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2026.3685449
“In both biological and artificial neural networks, operating near the edge of chaos has been associated with enhanced computational capabilities, including improved memory retention, sensitivity to inputs, and representational richness [44], [45], [46], [47].”
Intrinsic chaos control in cortical circuits: A minimal E-I-M rate model for primary visual cortex
Borjkhani, M., Sharif, M. A., & Borjkhani, H. (2026) · Journal of Computational Neuroscience · doi:10.1007/s10827-026-00938-5
“The “edge of chaos” hypothesis proposes that optimal computation occurs at the boundary between order and chaos (Bertschinger & Natschläger, 2004).”
Anonymous, Weerawongphrom, N., Williams-García, R. V., Beggs, J. M., & Ortíz, G. (2026) · Physical review. E · doi:10.1103/64b7-qb8p · Cited by 1
Quantitative Evaluation of Computational Capability in Complex Dynamical Systems
Ma, R., & Chen, T. X. (2026) · International Annual Conference on Complex Systems and Intelligent Science (CSIS-IAC) · doi:10.1109/csis-iac70275.2026.11585166
“These metrics are affected by multiple factors, including reservoir size ( N ), spectral radius (SR), connectivity, and activation function [3], [6], [7].”
Biochemical and gene regulatory dynamics as reservoir-like substrates for natural computing
Oosawa, C. (2026) · Biosystems · doi:10.1016/j.biosystems.2026.105895
Adaptive Dual-Path Reservoir Computing for Systemic Risk Early Warning via Connectedness ForecastingPreprint
Fang, Z., Gu, C., Liao, X., Gao, G., Wang, Z., Sun, T., Liu, Y., & Wang, J. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7170620
Avalanche assemblies in arbitrary connection topologies
Anonymous, Keßeböhmer, M., & Ernst, U. A. (2026) · Physical review. E · doi:10.1103/nl7d-16n8
Non-Hermitian quantum reservoir computing
Zhang, L., Wu, Y., Gao, Y., & Wang, C. (2026) · Quantum Science and Technology · doi:10.1088/2058-9565/ae9abe
“This dynamical crossover is relevant for reservoir computing because reservoir performance is often optimized near an intermediate regime between overly stable and overly unstable dynamics, commonly referred to as the edge of chaos [33, 34, 39, 48–50].”
Kinetic energy in random recurrent neural networks
Anonymous, & Huang, H. (2026) · Physical review. E · doi:10.1103/x15s-84n9 · Cited by 1
Garvey, M. D. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7428998
Alhawarat, M. O., Elsimary, H., & Aburrub, K. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7434435
Harnessing cavity-soliton chaos for reservoir computing in an injected laser with saturable absorber
Afridi, M. I., Ishtiaq, U., Faridi, W. A., & Alshehri, M. H. (2026) · Chaos Solitons & Fractals · doi:10.1016/j.chaos.2026.119112
The Role of Echo State and Fading Memory Properties in Reservoir Computing Systems
Smith, C. J., & Blain, J. (2026) · IEEE International Midwest Symposium on Circuits and Systems (MWSCAS) · doi:10.1109/mwscas67364.2026.11680701
“The effective SR in RC is often chosen close to unity to balance stability with temporal memory, since smaller values lead to fast state decay, whereas values near the stability boundary can support richer dynamics and longer memory [15], [25], [27].”
Zhang, S., Pan, Y., & Ge, X. (2026) · IEEE/CIC International Conference on Communications in China (ICCC) · doi:10.1109/iccc70295.2026.11680353
“This value, being close to unity, characterizes the emergence of 1 /f -like noise, which is commonly observed in complex systems near critical regimes [17].”
Parra, A. L. (2026) · Frontiers in Integrative Neuroscience · doi:10.3389/fnint.2026.1841714
“…a maximum near γ c and then declines at supracritical γ̂ values, consistent with tonic-high LC-mediated degradation of signal-to-noise (Aston-Jones and Cohen, 2005; Sara and Bouret, 2012) and with peaked-computation accounts of criticality (Bertschinger and Natschläger, 2004; Shew et al., 2009).”
InfluentialCuevas, E. M., & León, A. (2026) · Procedia Computer Science · doi:10.1016/j.procs.2026.08.189
Rajeev, R., Kamat, R., Kothapalli, M., Polavarapu, P., Chittimalla, S. K., & Potluri, L. K. M. (2026) · International Conference on Secure Information Systems and Technologies (ICSIST) · doi:10.1109/icsist70215.2026.11706135
“Recurrent networks can exhibit chaotic dynamics [12], [13].”
Lucken, C. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-39953-3_14
Lin, Z. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7629592
Vivancos, D., & Sanchez, J. (2026) · International Conference on Advanced Machine Learning and Data Science (AMLDS) · doi:10.1109/amlds69611.2026.11724215
“Operating near criticality has been associated with maximal dynamic range [9], [10], efficient real-time computation [11], and enhanced computational capacity across many types of neural models [12].”
InfluentialDesign Principles for Light‐Scattering Reservoir Computing at the Edge of Chaos
Kim, P. G., & Park, Y. (2026) · Advanced Computing · doi:10.1002/adco.70012
Metadata
- Authors
- Bertschinger, N., & Natschläger, T.
- Affiliations
- Graz University of Technology (AT)
- Software Competence Center Hagenberg (Austria) (AT)
- ORCID
- Bertschinger, Nils
- Venue
- Neural Computation
- Publisher
- MIT Press - Journals
- Published
- 2004-07-01
- Volume
- 16
- Issue
- 7
- Pages
- 1413-1436
- ISSN
- 0899-7667 (ISSN-L)
- 1530-888X
- OpenAlex
- W2057755457
- Cited by
- 833
- References
- 26
- Citations by year
- 2026: 45 · 2025: 73 · 2024: 61 · 2023: 48 · 2022: 66 · 2021: 73 · 2020: 52 · 2019: 53 · 2018: 47 · 2017: 48 · 2016: 26 · 2015: 33 · 2014: 29 · 2013: 33 · 2012: 27
- FWCI
- 13.53
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- Neural Computation (publishedVersion)
- PubMed (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Physical Sciences › Computer Science › Computational Theory and Mathematics › Cellular Automata and Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Reservoir Computing
- Keywords
- edge of chaos, recurrent neural networks, critical connectivity, threshold gates, random connectivity, real-time processing, dynamical systems
- MeSH terms
- Animals, Artificial Intelligence, Computer Simulation, Feedback, Memory, Models, Neurological, Time Factors, Neural Networks, Computer, Nonlinear Dynamics
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.