Dynamic instabilities from information annihilation in neuronal networks and human motor control

许多复杂自适应系统，包括神经网络，人类平衡行为和金融市场，表现出复杂的活动，其特征在于时空标度律。有趣的是，这些系统都具有相反影响的动态平衡。我们的目的是解释为什么这种平衡不会导致简单的均衡。特别是，我们将调查是否有一个单一的一般原则，其中的关键出现从一个有效的吸收信息，可以解释这些意见。我们将在两个关于运动控制（SP1）和神经网络（SP2）的子项目中研究这些问题：自适应运动控制（SP1）：我们以前的工作表明，信息湮灭不稳定性（IAI）可以解释自适应控制中的临界性。然而，在何种程度上相应的模型可以解释人类的运动控制在现实情况下是未知的。我们将进行实验来测试我们以前的模型的预测，研究我们的理论的适用范围，并进行后续的修改和扩展。此外，我们还将研究我们的理论对自适应控制理论、前向模型和电机控制策略的自组织的更普遍的影响。我们希望这个子项目产生新的见解，人类运动控制和提供重要的限制，为发展生物现实的神经网络models.Neuronal网络（SP2）：大脑的许多部分的运作接近一个状态，兴奋和抑制输入发送到神经元是密切平衡。造成这种平衡的机制及其功能意义尚不清楚。初步证据表明，特定的突触适应，实现输入平衡，诱导网络对意外输入的敏感性增加，并导致稀疏和预测代码。我们进一步怀疑，在递归网络中，平衡状态对应于有利于编码、信号传输和计算的特定操作点。这些假设中的每一个都将通过开发和分析生物学上合理的神经网络模型来研究。此外，动态后果和功能的平衡状态的好处将探讨范式网络类型，以及为选定的闭环传感器电机控制系统。使用数据驱动的方法，我们将确定这两个系统的差异和相似之处以及它们的动态，并在生物学上合理的约束条件下开发相应的模型。这将为IAI的一般理论的制定奠定基础。将在SP1、SP2中并行进行统一的、独立于系统的描述的识别。如果获得DFG批准，将与金融市场IAI相关项目密切合作（Felix Patzelt博士）。

Many Complex Adaptive Systems, including neuronal networks, human balancing behaviour, and financial markets, exhibit complex activity, which are characterised by spatio-temporal scaling laws. Interestingly, these systems all feature a dynamic balance of opposing influences. We aim to explain why this balancing does not result in simple equilibria. In particular, we will investigate whether a single general principle, by which criticality emerges from an efficient absorption of information, can account for these observations. We will work on these questions in two subprojects concerning motor control (SP1) and neuronal networks (SP2):Adaptive motor control (SP1): Our previous work shows that Information Annihilation Instability (IAI) can account for criticality in adaptive control. However, the extent to which the corresponding models can explain human motor control in realistic situations is not known. We will perform experiments to test predictions of our previous model, to investigate the range of applicability of our theory, and to perform succeeding modifications and extensions. Also, we will investigate the more general consequences of our theory for adaptive control theory, forward models, and self-organisation of motor-control policies. We expect this subproject to yield new insights into human motor control and to deliver important constraints for the development of biologically realistic neural network models.Neuronal networks (SP2): Many parts of the brain operate close to a state where exciting and inhibiting inputs sent to a neuron are closely balanced. The mechanisms causing this balance and its functional significance are not well understood. Preliminary evidence indicates that specific synaptic adaptations, realizing input balance, induce an increased sensitivity of networks to unexpected inputs and lead to sparse and predictive codes. We further suspect that in recurrent networks the balanced state corresponds to a specific point of operation that is beneficial for coding, signal transmission, and computation. Each of these hypotheses will be investigated by developing and analysing biologically plausible neuronal network models. Furthermore, dynamical consequences and functional benefits of the balance state will be explored for paradigmatic network types as well as for selected closed loop sensor-motor control systems. Using a data-driven approach, we will identify differences and similarities of the two systems as well as their dynamics and develop the respective models within biologically plausible constraints. This will lay ground for the formulation of a general theory of IAI. The identification of unifying, system-independent descriptions will be carried out in parallel in SP1, SP2. and, if approved by the DFG, in close collaboration with the related project on IAI in financial markets (by Dr. Felix Patzelt).