Real-Time Computing Without Stable States: A New Framework for Neural Computation Based on Perturbations

By Wolfgang Maass, Thomas Natschläger, Henry Markram. 2002

4,178 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Cognitive Neuroscience, Electrical and Electronic Engineering

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Cognitive Neuroscience

In Electrical and Electronic Engineering

Citations in the corpus

Deep learning in neural networks: An overview (2014, neutral): “In fact, some popular RNN algorithms restricted credit assignment to a single step backwards (Elman, 1990; Jordan, 1986, 1997), also inmore recent studies (Jaeger, 2001, 2004; Maass et al., 2002).”
Capacity and Trainability in Recurrent Neural Networks (2016, background): “Previous work on short term memory traces explores the tradeoffs between memory ﬁdelity and duration, for the case that a new input is presented to the RNN at every time step (Jaeger & Haas, 2004; Maass et al., 2002; White et al., 2004; Ganguli et al., 2008; Charles et al., 2014).”
Toward an Integration of Deep Learning and Neuroscience (2016, background): “For example, “liquid-” (Maass et al., 2002) or “echo-state machines” (Jaeger and Haas, 2004) are randomly connected recurrent networks that form a basis set (also known as a “reservoir”) of random ﬁlters, which can be harnessed for learning with tunable readout weights.”
Learn more by training less: systematicity in sentence processing by recurrent networks (2006, background): “Indeed, independently from each other, Jaeger (2003) and Maass et al. (2002) proposed a new type of recurrent neural network based on the idea that the recurrent, dynamical part of the network does not require training.”
At the Edge of Chaos: How Cerebellar Granular Layer Network Dynamics Can Provide the Basis for Temporal Filters (2015, background): “This however was almost exclusively done for generic networks where neurons are always connected by negative and positive weights [27] or include both mutual exciting and inhibiting neurons in the randomly connected reservoir networks [28].”
Introduction to spiking neural networks: Information processing, learning and applications. (2011, methodology): “Yet, some attempts to harvest precise spiking timing in recurrent networks have been made for example by exploring the idea of reservoir computation (Maass et al. 2002a, Ponulak and Kasinski 2010).”
Physical reservoir computing—an introductory perspective (2020, background): “Accordingly, several reservoir settings implement brain-inspired connections [134] or simply implement the neighboring connections [4], introducing a spatial dimension that is not random at all.”
Artificial Neural Networks-Based Machine Learning for Wireless Networks: A Tutorial (2017, methodology): “selected inhibitory neurons and 80% excitatory neurons. Connections from input to liquid: for every pair of input signal and liquid neuron there is a certain chance of being connected, e.g., 30% in [121]. Connections between neurons in the liquid: The connections between the neurons are allocated ”
Backpropagation algorithms and Reservoir Computing in Recurrent Neural Networks for the forecasting of complex spatiotemporal dynamics (2019, methodology): “This technique has been proposed in the context of Echo-State-Networks (ESNs) (Jaeger & Haas, 2004) and Liquid State Machines with spiking neurons (LSM) (Maass, Natschläger, & Markram, 2002).”
A Dynamical System-based Approach to Modeling Stable Robot Control Policies via Imitation Learning (2012, neutral): “, 1985), Reservoir Computing (Lukosevicius & Jaeger, 2009; Maass et al., 2002), etc.”
Adaptive Nonlinear System Identification with Echo State Networks (2002, background): “A similar idea has recently been independently investigated in a more biologically oriented setting under the name of "liquid state networks" [8] [9].”
Synaptic Plasticity Dynamics for Deep Continuous Local Learning (DECOLLE) (2018, methodology): “This is achieved using a local rate-based cost function reminiscent of readout mechanisms used in liquid state machines [15], but where the readout is performed over a fixed random combination of the neuron outputs.”
Minimum Complexity Echo State Network (2011, background): “Liquid state machines (LSMs) [33] also mostly have a linear readout and the reservoirs are driven by the dynamics of a set of coupled spiking neuron models.”
High-Speed Photonic Reservoir Computing Using a Time-Delay-Based Architecture: Million Words per Second Classification (2017, background): “Such auto-oscillations are considered to prevent an important required property for reservoir dynamics, referred to as the fading memory property [3].”
Deep neural networks: a new framework for modelling biological vision and brain information processing (2015, background): “In an echo-state network (Jaeger 2001, see also Maass et al. 2002, for a similar model with spiking dynamics), for example, the sequence of input patterns is fed into a set of hidden units that are sparsely and randomly connected.”
Towards a theoretical foundation for morphological computation with compliant bodies (2011, methodology): “For a detailed proof, we refer to Theorem 1 in Maass et al. (2002) and Theorem 3.1 in Maass and Sontag (2000) and their corresponding proofs.”
Pattern Recognition in a Bucket (2003, background): “Maass et al [1] have produced a model called the Liquid State Machine (LSM) which hypothesises that a cortical microcolumn consists of stereotyped recurrent circuits of integrate-and-fire neurons connected randomly according to a few parameters.”