Adaptive Central Pattern Generator based on Spiking Neural Networks
暂无分享,去创建一个
In this paper we present a new, adaptive model of the Central Pattern Generator (CPG) [1] based on Spiking Neural Networks (SNN) [2]. The model has the ability to learn the desired rhythmic patterns from demonstration. Central Pattern Generators (CPG) play the principle role in such processes as locomotion, breathing or heart-beating of animals. For this reason they are of great interest for scientists. Many models of CPGs have already been proposed [3]. However, most of these models are suitable only for the individually designated tasks. It is desirable to construct the generic CPG model with the universal ability to learn the task from the given, desired motor-patterns. An interesting approach to such a programmable CPG has been proposed in [4]. In that approach a CPG model has been based on Hopf oscillators. Here, we introduce a model of an adaptive CPG based on SNN. Our model can learn the desired motorpattern to be generated in each cycle of CPG operation. The learning process is performed according to the Remote Supervised Method (ReSuMe) 1 introduced in [2]. We describe our approach in an example of a CPG model with 2 outputs. This corresponds e.g. to the CPG generating the alternating flexor-extensor activity responsible for limb coordination during locomotion. Our model can be divided into two functional parts: rhythm and pattern generator [1] (Fig.1). The rhythm generator produces the basic oscillations and controls their frequency. This part consists of 2 reciprocally connected networks (A, B) of spiking neurons. The networks have the sparse recurrent organization. In our recent study we are concerned only with reproducing the desired shape of the motor-patterns. Hence, we assume that the parameters of the rhythm generator, such as the frequency of oscillations, the duty cycle and the phase relationship are not modified
[1] E. Marder,et al. Central pattern generators and the control of rhythmic movements , 2001, Current Biology.
[2] Scott L. Hooper. Central Pattern Generators , 2001 .
[3] A. Ijspeert,et al. From Dynamic Hebbian Learning for Oscillators to Adaptive Central Pattern Generators , 2005 .
[4] F. Ponulak. ReSuMe-New Supervised Learning Method for Spiking Neural Networks , 2005 .