Echo State Networks used for Motor Control

Spiking Neuron Networks (SNNs) are often referred to as the 3 rd generation of neural networks. Highly inspired from natural computing in the brain and recent advances in neurosciences, they derive their strength and interest from an accurate modeling of synaptic interactions between neurons, taking into account the time of spike firing. SNNs overcome the computational power of neural networks made of threshold or sigmoidal units. Based on dynamic event-driven processing, they open up new horizons for developing models with an exponential capacity of memorizing and a strong ability to fast adaptation. Today, the main challenge is to discover efficient learning rules that might take advantage of the specific features of SNNs while keeping the nice properties (general-purpose, easy-to-use, available simulators, etc.) of traditional connectionist models. This chapter relates the history of the "spiking neuron" in Section 1 and summarizes the most currently-in-use models of neurons and synaptic plasticity in Section 2. The computational power of SNNs is addressed in Section 3 and the problem of learning in networks of spiking neurons is tackled in Section 4, with insights into the tracks currently explored for solving it. Finally, Section 5 discusses application domains, implementation issues and proposes several simulation frameworks.

show abstract

“…[142]), prediction in the context of wireless telecommunications (e.g. [73]) or neuromorphic approaches to rehabilitation, in medecine (e.g.…”

Section: Pattern Recognition With Snnsmentioning

confidence: 99%

Computing with Spiking Neuron Networks

Paugam-Moisy

Bohté

2012

Handbook of Natural Computing

207

114

View full text Add to dashboard Cite

show abstract

“…In some sense, an ESN can be seen as a universal dynamical system approximator, which linearly combines the elementary dynamics contained in the reservoir [15]. ESN have been shown to perform surprisingly well in the context of supervised learning, in particular for problems of prediction of times series, though it has also been successfully used in the context of (supervised) robot control learning [16] (also see [1] for an overview of ESN applications). Figure 2 illustrates the classic learning protocol for ESN using simple linear regression as learning algorithm.…”

Section: Echo State Networkmentioning

confidence: 99%

Using echo state networks for robot navigation behavior acquisition

Hartland

Bredèche

2007

2007 IEEE International Conference on Robotics and Biomimetics (ROBIO)

View full text Add to dashboard Cite

Abstract-Robot Behavior Learning by Demonstration deals with the ability for a robot to learn a behavior from one or several demonstrations provided by a human teacher, possibly through tele-operation or imitation. This implies controllers that can address both (1) the feature selection problem related to a great amount of mostly irrelevant sensory data and (2) dealing with temporal sequences of demonstrations. Echo State Networks [10] have been proposed recently for time series prediction and have been shown to perform remarkably well on this kind of data. In this paper, we introduce ESN to robot behavior acquisition in the scope of a mobile robot performing navigation tasks. ESN actually show comparable and even better performance with that of other algorithms from the literature in similar experimental conditions. Moreover, some properties regarding dynamics of ESN in the context of learning by demonstration are investigated.

show abstract

“…Furthermore, the training process corresponds to solving a linear regression problem and thus, does not suffer from problems such as slow convergence or sub-optimality that are inherent in most gradient-based methods used for training of RNNs. Due to this advantage, ESNs have found various applications such as predicting chaotic and nonlinear systems [23][24][25], motor speed control [26], online classification of visual tasks [27], learning grammatical structures [28], automatic speech recognition [29], control of shape memory alloys [30], forecasting short-term electric load [31], nonlinear adaptive filtering of complex signals [32], and modeling and control of pneumatic artificial muscles [33].…”

Section: Introductionmentioning

confidence: 99%