Pavel V. Kuptsov

Pavel V. Kuptsov, Nataliya V. Stankevich, Elmira R. Bagautdinova

We consider Hodgkin-Huxley-type model that is a stiff ODE system with two fast and one slow variables. For the parameter ranges under consideration the original version of the model has unstable fixed point and the oscillating attractor that demonstrates bifurcation from bursting to spiking dynamics. Also a modified version is considered where the bistability occurs such that an area in the parameter space appears where the fixed point becomes stable and coexists with the bursting attractor. For these two systems we create artificial neural networks that are able to reproduce their dynamics. The created networks operate as recurrent maps and are trained on trajectory cuts sampled at random parameter values within a certain range. Although the networks are trained only on oscillatory trajectory cuts, it also discover the fixed point of the considered systems. The position and even the eigenvalues coincide very well with the fixed point of the initial ODEs. For the bistable model it means that the network being trained only on one brunch of the solutions recovers another brunch without seeing it during the training. These results, as we see it, are able to trigger the development of new approaches to complex dynamics reconstruction and discovering. From the practical point of view reproducing dynamics with the neural network can be considered as a sort of alternative method of numerical modeling intended for use with contemporary parallel hard- and software.

Pavel V. Kuptsov, Nataliya V. Stankevich

This paper examines the reconstruction of a family of dynamical systems with neuromorphic behavior using a single scalar time series. A model of a physiological neuron based on the Hodgkin-Huxley formalism is considered. Single time series of one of its variables is shown to be enough to train a neural network that can operate as a discrete time dynamical system with one control parameter. The neural network system is created in two steps. First, the delay-coordinate embedding vectors are constructed form the original time series and their dimension is reduced with by means of a variational autoencoder to obtain the recovered state-space vectors. It is shown that an appropriate reduced dimension can be determined by analyzing the autoencoder training process. Second, pairs of the recovered state-space vectors at consecutive time steps supplied with a constant value playing the role of a control parameter are used to train another neural network to make it operate as a recurrent map. The regimes of thus created neural network system observed when its control parameter is varied are in very good accordance with those of the original system, though they were not explicitly presented during training.

Pavel V. Kuptsov, Anna V. Kuptsova, Nataliya V. Stankevich

We suggest a universal map capable to recover a behavior of a wide range of dynamical systems given by ODEs. The map is built as an artificial neural network whose weights encode a modeled system. We assume that ODEs are known and prepare training datasets using the equations directly without computing numerical time series. Parameter variations are taken into account in the course of training so that the network model captures bifurcation scenarios of the modeled system. Theoretical benefit from this approach is that the universal model admits using common mathematical methods without needing to develop a unique theory for each particular dynamical equations. Form the practical point of view the developed method can be considered as an alternative numerical method for solving dynamical ODEs suitable for running on contemporary neural network specific hardware. We consider the Lorenz system, the Roessler system and also Hindmarch-Rose neuron. For these three examples the network model is created and its dynamics is compared with ordinary numerical solutions. High similarity is observed for visual images of attractors, power spectra, bifurcation diagrams and Lyapunov exponents.

Pavel V. Kuptsov, Anna V. Kuptsova

Dimension of an inertial manifold for a chaotic attractor of spatially distributed system is estimated using autoencoder neural network. The inertial manifold is a low dimensional manifold where the chaotic attractor is embedded. The autoencoder maps system state vectors onto themselves letting them pass through an inner state with a reduced dimension. The training processes of the autoencoder is shown to depend dramatically on the reduced dimension: a learning curve saturates when the dimension is too small and decays if it is sufficient for a lossless information transfer. The smallest sufficient value is considered as a dimension of the inertial manifold, and the autoencoder implements a mapping onto the inertial manifold and back. The correctness of the computed dimension is confirmed by its remarkable coincidence with the one obtained as a number of covariant Lyapunov vectors with vanishing pairwise angles. These vectors are called physical modes. Unlike never having zero angles residual ones they are known to span a tangent subspace for the inertial manifold.