Adaptive Learning Of Echo State Networks Based On Neural Plasticity Mechanisms

Based on the experimental findings of neurobiology and reasonable computational hypotheses,computational modeling of neural plasticity mechanisms has been investigated to improve the learning performance of artificial neural networks.Simulating the working processes of biological neural mechanisms by using computational modeling can help human to better understand the internal working mechanism and biological computing process of neural systems in the process of information representation,processing and storage.However,existing computational models of neural plasticity also suffer from several limitations,such as the instability of synaptic plasticity learning process,poor learning performance due to single neural plasticity rule,and the lack of adequate understanding the underlying mechanisms of biological neural plasticity.Based on the findings in computational neuroscience and neurobiology studies,this dissertation focuses developing new computational plasticity rules with their application to echo state networks for solving various synthetic and realworld regression problems.The main new contributions of this PhD thesis include:(1)Existing synaptic plasticity rules for optimizing the connections between neurons within the reservoir of echo state networks(ESNs)remain to be global in that the same type of plasticity rule with the same parameters is applied to all neurons.However,this is biologically implausible and practically inflexible for learning the structures in the input signals,thereby limiting the learning performance of echo state networks.In this dissertation,we propose to use local plasticity rules that allow different neurons to use different types of plasticity rules as well as different parameters,which is achieved by optimizing the parameters of the local plasticity rules using the evolution strategy with covariance matrix adaptation.We show that evolving neural plasticity will result in a synergistic learning of different plasticity rules,which plays an important role in improving the learning performance.Meanwhile,we show that the local plasticity rules can effectively alleviate synaptic interferences in learning the structure in sensory inputs.The proposed local plasticity rules are compared with a number of the state-of-the-art echo state network models and the canonical echo state network using a global plasticity rule on a set of widely used prediction and classification benchmark problems to demonstrate its competitive learning performance.(2)Intrinsic plasticity,as a biologically inspired unsupervised learning rule,is used for adapting the intrinsic excitability of the reservoir neurons.Existing intrinsic plasticity rules can only select a set of fixed rule parameters for the whole reservoir neurons,which affects the learning performance due to the lack of flexibility in providing intrinsic plasticity.Here,we present an echo state network with local intrinsic plasticity rule built by different reservoir neurons which can adopt the intrinsic plasticity rule with different rule parameters to adjust its intrinsic excitability.And the covariance matrix adaptation evolution strategy is used to search and select the rule parameters corresponding to different reservoir neurons.Compared with several state-of-the-art ESN models and an ESN with the global plasticity rule,the proposed local intrinsic plasticity rule is able to achieve much better performance in some benchmark prediction tasks.(3)Synaptic plasticity and intrinsic plasticity,as two of the most common neural plasticity mechanisms,occur in all neural circuits throughout life.Neurobiological studies indicated that the interplay between synaptic and intrinsic plasticity contributes to the adaptation of the nervous system to different synaptic input signals.However,most existing computational models of neural plasticity consider these two plasticity mechanisms separately,which is biologically implausible.To address this issue,this this thesis proposes a synergistic plasticity learning rule to adapt the reservoir connections in ESNs,which not only takes into account the regulation of synaptic weights,but also considers the adjustment of neuronal intrinsic excitability.The proposed synergetic plasticity rule is verified on a number of prediction and classification benchmark problems and our empirical results demonstrate that the ESN with synergistic plasticity learning rule performs much better than the state-of-the-art ESN models,and an ESN with a single neural plasticity rule.(4)Existing studies on computational modeling of neural plasticity have mostly focused on synaptic plasticity.However,the regulation process of alone synaptic plasticity resulting the unstable learning behavior still existed in some machine learning tasks.In this work,a novel structural synaptic plasticity learning rule is constructed to adjust the connectioning weights and the addition/removal of neurons within the reservoir,which can overcome the instability of the synaptic plasticity learning process.During the experiments we also find that a few large and steady connection weights could last a long time in constantly changing network structure.Finally,the ESN with structural synaptic plasticity rule is compared with several state-of-the-art ESN models and an ESN with the synaptic plasticity rule in some benchmark prediction tasks.The comparstive results show that the ESN with the structural synaptic plasticity learning rule can obtain much better performance.