An overview of reservoir computing: theory, applications and implementations
Ghent University Academic Bibliography (Ghent University)Published 1 January 2007Open access
Benjamin Schrauwen, David Verstraeten, Jan M. Van Campenhout
Citations327
Generate an AI Snapshot to get a quick, structured summary of this paper.
Study Snapshot
ObjectiveStudy objective
MethodsResearch methodology
PopulationPopulation studied
Sample sizeSample sizes
OutcomesStudy outcomes here
ResultsStudy results comes here
LimitationsResearch study limitations comes here
A concise AI-generated summary of the paper will appear here once you click Generate AI Snapshot.
TL;DR
This tutorial will give an overview of current research on theory, applica- tion and implementations of Reservoir Computing, which makes it possible to solve complex tasks using just linear post-processing techniques.
Abstract
Abstract. Training recurrent neural networks is hard. Recently it has however been discovered that it is possible to just construct a random recurrent topology, and only train a single linear readout layer. State-ofthe-art performance can easily be achieved with this setup, called Reservoir Computing. The idea can even be broadened by stating that any high dimensional, driven dynamic system, operated in the correct dynamic regime can be used as a temporal ‘kernel ’ which makes it possible to solve complex tasks using just linear post-processing techniques. This tutorial will give an overview of current research on theory, application and implementations of Reservoir Computing. 1
Keywords
Computer ScienceNeuroscienceEngineering
Neural ComputationLong Short-Term Memory
98,079 Citations1997Sepp Hochreiter, Jürgen Schmidhuber
A novel, efficient, gradient based method called long short-term memory (LSTM) is introduced, which can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units.
Proceedings of the National Academy of SciencesNeural networks and physical systems with emergent collective computational abilities.
19,349 Citations1982J. J. Hopfield
A model of a system having a large number of simple equivalent components, based on aspects of neurobiology but readily adapted to integrated circuits, produces a content-addressable memory which correctly yields an entire memory from any subpart of sufficient size.
Learning Internal Representations by Error Propagation
16,235 Citations1985David E. Rumelhart, Geoffrey E. Hinton +1 more
The MIT Press eBooksLearning with Kernels
9,548 Citations2001Bernhard Schölkopf, Alexander J. Smola
Learning with Kernels provides an introduction to SVMs and related kernel methods that provide all of the concepts necessary to enable a reader equipped with some basic mathematical knowledge to enter the world of machine learning using theoretically well-founded yet easy-to-use kernel algorithms.
Proceedings of the IEEEBackpropagation through time: what it does and how to do it
4,904 Citations1990Paul J. Werbos
This paper first reviews basic backpropagation, a simple method which is now being widely used in areas like pattern recognition and fault diagnosis, and describes further extensions of this method, to deal with systems other than neural networks, systems involving simultaneous equations or true recurrent networks, and other practical issues which arise with this method.
Neural ComputationReal-Time Computing Without Stable States: A New Framework for Neural Computation Based on Perturbations
4,037 Citations2002Wolfgang Maass, Thomas Natschläger +1 more
A new computational model for real-time computing on time-varying input that provides an alternative to paradigms based on Turing machines or attractor neural networks, based on principles of high-dimensional dynamical systems in combination with statistical learning theory and can be implemented on generic evolved or found recurrent circuitry.
ScienceHarnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication
3,792 Citations2004Herbert Jaeger, Harald Haas
A method for learning nonlinear systems, echo state networks (ESNs), which employ artificial recurrent neural networks in a way that has recently been proposed independently as a learning mechanism in biological brains is presented.
Neural NetworksAn experimental unification of reservoir computing methods
1,131 Citations2007D. Verstraeten, Benjamin Schrauwen +2 more
Three different uses of a recurrent neural network as a reservoir that is not trained but instead read out by a simple external classification layer are compared and a new measure for the reservoir dynamics based on Lyapunov exponents is introduced.
Neural NetworksApproximation of dynamical systems by continuous time recurrent neural networks
962 Citations1993Ken-ichi Funahashi, Yuichi Nakamura
It is proved that any finite time trajectory of a given n-dimensional dynamical system can be approximately realized by the internal state of the output units of a continuous time recurrent neural network with n output units, some hidden units, and an appropriate initial condition.
Neural NetworksOptimization and applications of echo state networks with leaky- integrator neurons
927 Citations2007Herbert Jaeger, Mantas Lukoševičius +2 more
Stability conditions are presented, a stochastic gradient descent method is introduced and a leaky-integrator ESNs are demonstrated for learning very slow dynamic systems and replaying the learnt system at different speeds.
Nature reviews. NeuroscienceThe other side of the engram: experience-driven changes in neuronal intrinsic excitability
884 Citations2003Wei Zhang, David J. Linden
IEEE Transactions on Neural NetworksLSTM recurrent networks learn simple context-free and context-sensitive languages
726 Citations2001Felix A. Gers, E. Schmidhuber
Long short-term memory (LSTM) variants are also the first RNNs to learn a simple context-sensitive language, namely a(n)b( n)c(n).
Adaptive Nonlinear System Identification with Echo State Networks
532 Citations2002Herbert Jaeger
An online adaptation scheme based on the RLS algorithm known from adaptive linear systems is described, as an example, a 10-th order NARMA system is adaptively identified.
IEEE Transactions on Neural NetworksNeurocontrol of nonlinear dynamical systems with Kalman filter trained recurrent networks
525 Citations1994G.V. Puskorius, L.A. Feldkamp
These simulations suggest that recurrent controller networks trained by Kalman filter methods can combine the traditional features of state-space controllers and observers in a homogeneous architecture for nonlinear dynamical systems, while simultaneously exhibiting less sensitivity than do purely feedforward controller networks to changes in plant parameters and measurement noise.
IEEE Transactions on Neural NetworksNew results on recurrent network training: unifying the algorithms and accelerating convergence
498 Citations2000Amir F. Atiya, A.G. Parlos
An on-line version of the proposed algorithm, which is based on approximating the error gradient, has lower computational complexity in computing the weight update than the competing techniques for most typical problems and reaches the error minimum in a much smaller number of iterations.
ScienceTemporal Information Transformed into a Spatial Code by a Neural Network with Realistic Properties
467 Citations1995Dean V. Buonomano, Michael M. Merzenich
It is demonstrated that known time-dependent neuronal properties enable a network to transform temporal information into a spatial code in a self-organizing manner, with no need to assume a spectrum of time delays or to custom-design the circuit.
IEEE Transactions on Neural NetworksAn application of recurrent nets to phone probability estimation
445 Citations1994A.J. Robinson
Recognition results are presented for the DARPA TIMIT and Resource Management tasks, and it is concluded that recurrent nets are competitive with traditional means for performing phone probability estimation.
Neural ComputationAnalysis and Design of Echo State Networks
312 Citations2006Mustafa C. Ozturk, Dongming Xu +1 more
A functional space approximation framework is presented to better understand the operation of ESNs and an information-theoretic metric, the average entropy of echo states, is proposed to assess the richness of the ESN dynamics.
Lecture notes in computer sciencePattern Recognition in a Bucket
260 Citations2003Chrisantha Fernando, Sampsa Sojakka
This paper demonstrates that the waves produced on the surface of water can be used as the medium for a “Liquid State Machine” that pre-processes inputs so allowing a simple perceptron to solve the XOR problem and undertake speech recognition.
Information Processing LettersIsolated word recognition with the Liquid State Machine: a case study
256 Citations2005D. Verstraeten, Benjamin Schrauwen +2 more
A case study of the performance of the Liquid State Machine based on a recurrent spiking neural network by applying it to a well known and well studied problem: speech recognition of isolated digits and finds the biologically most realistic encoding performs far better than more conventional methods.
Neural ComputationTraining Recurrent Networks by Evolino
254 Citations2007Jürgen Schmidhuber, Daan Wierstra +2 more
Cerebral CortexA Statistical Analysis of Information-Processing Properties of Lamina-Specific Cortical Microcircuit Models
228 Citations2006Stefan Haeusler, Wolfgang Maass
It is concluded that computer simulations of detailed lamina-specific cortical microcircuit models provide new insight into computational consequences of anatomical and physiological data.
Backpropagation-decorrelation: online recurrent learning with O(N) complexity
202 Citations2005Jochen J. Steil
A new learning rule for fully recurrent neural networks is introduced which combines important principles: one-step backpropagation of errors and the usage of temporal memory in the network dynamics by means of decorrelation of activations.
Neural NetworksDecoupled echo state networks with lateral inhibition
190 Citations2007Yanbo Xue, Le Yang +1 more
In a real-life prediction task using noisy sea clutter data, both schemes exhibit higher prediction accuracy and successful design ratio than a conventional ESN with a single reservoir.
Neural NetworksOnline reservoir adaptation by intrinsic plasticity for backpropagation–decorrelation and echo state learning
165 Citations2007Jochen J. Steil
It is shown experimentally that a biologically motivated learning rule based on neural intrinsic plasticity can drive the neurons' output activities to approximate exponential distributions and implement sparse codes in the reservoir.
IEEE Transactions on Neural NetworksA tighter bound for the echo state property
156 Citations2006Michael R. Buehner, Peter M. Young
The stability bounds presented here could aid in the design of echo state networks that would be applicable to control applications where stability is required.
Information and ComputationThe Dynamic Universality of Sigmoidal Neural Networks
119 Citations1996Joe Kilian, Hava T. Siegelmann
The techniques can be applied to a much more general class of “sigmoidal-like” activation functions, suggesting that Turing universality is a relatively common property of recurrent neural network models.
Lecture notes in computer scienceBiologically Plausible Speech Recognition with LSTM Neural Nets
110 Citations2004Alex Graves, Douglas Eck +2 more
It is concluded that LSTM should be further investigated as a biologically plausible basis for a bottom-up, neural net-based approach to speech recognition.
Journal of Physiology-ParisFading memory and kernel properties of generic cortical microcircuit models
95 Citations2004Wolfgang Maass, Thomas Natschläger +1 more
This article proposes to analyze circuits of spiking neurons in terms of their roles as analog fading memory and non-linear kernels, rather than as implementations of specific computational operations and algorithms.
Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005.Reservoir riddles: suggestions for echo state network research (extended abstract)
95 Citations2006Herbert Jaeger
This contribution discusses phenomena related to poor learning performance and suggests research directions to understand the reservoir dynamics in terms of a dynamical representation of the task's input signals.
Neural ComputationSimple Recurrent Networks Learn Context-Free and Context-Sensitive Languages by Counting
85 Citations2001Paul Rodriguez
A range of language tasks are shown in which an SRN develops solutions that not only count but also copy and store counting information, demonstrating how SRNs may be an alternative psychological model of language or sequence processing.
Infoscience (Ecole Polytechnique Fédérale de Lausanne)A Model for Real-Time Computation in Generic Neural Microcircuits
77 Citations2002Wolfgang Maass, Thomas Natschläger +1 more
A new computational model is proposed that is based on principles of high dimensional dynamical systems in combination with statistical learning theory and can be implemented on generic evolved or found recurrent circuitry.
Echo State Networks used for Motor Control
75 Citations2006M. Salmen, Paul G. Plöger
This paper validates for the first time the predicted superior simulation results by physical experiments that prove that the ESN controller shows a slightly better control quality as a PID controller with respect to various important error norms from control theory.
Minimum mean squared error time series classification using an echo state network prediction model
62 Citations2006Mark D. Skowronski, J.G. Harris
The ESN offers an attractive alternative to the HMM because of the ESN's simple train procedure, low computational requirements, and inherent ability to model the dynamics of the signal under study.
Is there a Liquid State Machine in the Bacterium Escherichia Coli?
60 Citations2007Ben Jones, Dov J. Stekel +2 more
Coupling of the GRN with signal transduction networks possessing cross-talk, and with metabolic networks is expected to increase the extent of non-autonomous recurrence and hence to facilitate enhanced LSM properties.
Synergies between Intrinsic and Synaptic Plasticity in Individual Model Neurons
55 Citations2004Jochen Triesch
A new intrinsic plasticity mechanism for a continuous activation model neuron based on low order moments of the neuron's firing rate distribution is proposed and the neuron is shown to discover sparse directions in the input.
NeurocomputingOnline stability of backpropagation–decorrelation recurrent learning
49 Citations2006Jochen J. Steil
A stability analysis based on nonlinear feedback theory for the recently introduced backpropagation-decorrelation (BPDC) recurrent learning algorithm which adapts only the output weights of a possibly large network and therefore can learn in O(N).
Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005.Modeling reward functions for incomplete state representations via echo state networks
44 Citations2005Keith Bush, Charles W. Anderson
This research demonstrates that the ESN architecture represents the Q-function of the MSD system given incomplete state information as well as current feed forward neural networks given either perfect state or a temporally-windowed, incomplete state vector.
Lecture notes in computer scienceMovement Generation and Control with Generic Neural Microcircuits
43 Citations2004Prashant Joshi, Wolfgang Maass
This approach provides methods for also using the “embodiment of computation”, i.e. the inherent dynamics and spatial structure of neural circuits, for the design of robot movement controllers.
Lecture notes in computer scienceA New Approach towards Vision Suggested by Biologically Realistic Neural Microcircuit Models
42 Citations2002Wolfgang Maass, Robert Legenstein +1 more
This approach supports parallel real-time processing of time-varying visual inputs for diverse tasks, since different readouts can be trained to extract concurrently from the same microcircuit completely different information components.
The MIT Press eBooksTemporal dynamics of information content carried by neurons in the primary visual cortex
41 Citations2007Danko Nikolić, Stefan Haeusler +2 more
It is found that information is extractable and that it even lasts for several hundred milliseconds after the stimulus has been removed, suggesting nonlinear properties of cortical representations in a fast sequence of stimulus presentation.
Reservoir Computing with Stochastic Bitstream Neurons
35 Citations2005David Verstraeten, Benjamin Schrauwen +1 more
A type of neuron is introduced that communicates using stochastic bitstreams instead of fixed-point values, which drastically simplifies the hardware implementation of arithmetic operations such as addition, the nonlinearity and multiplication on FPGA.
Edge of Chaos Computation in Mixed-Mode VLSI - A Hard Liquid
34 Citations2004Felix Schürmann, Karlheinz Meier +1 more
Results of a software implementation of a hardware implemented neural network are reproduced showing that the performance peaks when the network exhibits dynamics at the edge of chaos.
Learning Mappings in Brain Machine Interfaces with Echo State Networks
31 Citations2005Y.N. Rao, Sung-Phil Kim +6 more
E echo state networks (ESN) are proposed to be used to transform the neuronal firing activity into a higher dimensional space and then derive an optimal sparse linear mapping in the transformed space to match the hand position.
Learning to ground fact symbols in behavior-based robots
25 Citations2002Joachim Hertzberg, Herbert Jaeger +1 more
It is demonstrated that analogs of chronicle definitions can be learned using an echo state network, which is a particular type of recurrent neural network and provides the robot programmer an alternative tool for getting recent symbolic situation fact hypotheses.
Lecture notes in computer scienceShort Term Memory and Pattern Matching with Simple Echo State Networks
24 Citations2005Georg Fette, Julian Eggert
A simplified version of ESN which is called Simple Echo State Network (SESN) which exhibits good results in memory capacity and pattern matching tasks and which allows a better understanding of the capabilities and restrictions of ESNs.
Lecture notes in computer scienceMemory in Backpropagation-Decorrelation O(N) Efficient Online Recurrent Learning
23 Citations2005Jochen J. Steil
This work considers regularization methods to improve the recently introduced backpropagation-decorrelation (BPDC) online algorithm for O(N) training of fully recurrent networks and proposes several ways to introduce memory in the algorithm for regularization.
Constructing deterministic finite-state automata in sparse recurrent neural networks
19 Citations2002Christian W. Omlin, C. Lee Giles
An algorithm for encoding deterministic finite-state automata in sparse recurrent neural networks with sigmoidal discriminant functions and second-order weights and the authors prove that for particular weight strength values the regular languages accepted by DFAs and the constructed networks are identical.
IEEE International Conference on Neural Networks/IEEE ... International Conference on Neural NetworksCompact hardware for real-time speech recognition using a Liquid State Machine
18 Citations2007Benjamin Schrauwen, Michiel D’Haene +2 more
This work presents an application driven digital hardware exploration where a realtime, isolated digit speech recognition using a Liquid State Machine (a recurrent neural network of spiking neurons where only the output layer is trained) is implemented.
Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005.Computing with transiently stable states
18 Citations2006Mustafa C. Ozturk, José C. Prı́ncipe
A novel computational mode for nonlinear systems with sigmoidal nonlinearity, which does not require global stability, and although the autonomous system is unstable, the input signal forces the system dynamics to become "transiently stable".
Lecture notes in computer scienceVelocity Control of an Omnidirectional RoboCup Player with Recurrent Neural Networks
17 Citations2006Mohamed Oubbati, Michael Schanz +2 more
A recurrent neural network is used to develop a dynamic controller for mobile robots that has the advantage of the control approach is that no knowledge about the robot model is required.
IFAC Proceedings VolumesFinding good Echo State Networks to control an underwater robot using evolutionary computations
14 Citations2004Tijn van der Zant, Vlatko Bečanović +3 more
Two heuristics, evolutionary algorithms and evolutionary strategies, are compared and the proposed method outperforms standard ones like ARX and backpropagation networks on the data from the Twin-Burger underwater robot.
Lecture notes in computer scienceGenetic Regulatory Systems Modeled by Recurrent Neural Network
9 Citations2004Xianhua Dai
Based on the approximation capability of the ARNN, the proposed model can be used to express and analyze the genetic regulatory systems of genetic components, even the large-scale genetic systems.
Publikationsdatenbank der Fraunhofer-Gesellschaft (Fraunhofer-Gesellschaft)Echo state networks for mobile robot modeling and control
4 Citations2004Paul G. Plöger, A. Arghir +2 more