GradMDM: Adversarial Attack on Dynamic Networks.

Recurrent neural networks unlike feed-forward networks are able to process inputs with time context. The key role in this process is played by the dynamics of the network, which transforms input data to the recurrent layer states. Several authors have described and analyzed dynamics of small sized recurrent neural networks with two or three hidden units. In our work we introduce techniques that allow to visualize and analyze the dynamics of large recurrent neural networks with dozens units, reveal both stable and unstable points (attractors and saddle points), which are important to understand the principles of successful task processing. As a practical example of this approach, dynamics of the simple recurrent network trained by two different training algorithms on context-free language anbnwas studied.

A recent study (Chiyomaru and Takemoto 2022 Phys. Rev. E 106 014301) considered adversarial attacks conducted to distort voter model dynamics in networks. This method intervenes in the interaction patterns of individuals and induces them to be in a target opinion state through a small perturbation ε. In this study, we investigate adversarial attacks on voter dynamics in random networks of finite size n. The exit probability P +1 to reach the target absorbing state and the mean time τ n to reach consensus are analyzed in the mean-field approximation. Given ε > 0, the exit probability P +1 converges asymptotically to unity as n increases. The mean time τ n to reach consensus scales as for homogeneous networks with a large finite n. By contrast, it scales as for heterogeneous networks with a large finite n, where µ 1 and µ 2 represent the first and second moments of the degree distribution, respectively. Moreover, we observe the crossover phenomenon of τ n from a linear scale to a logarithmic scale and find above which the state of all nodes becomes the target state in logarithmic time. Here, α = 1 for homogeneous networks and for scale-free networks with a degree exponent .

In this paper a Dynamic Neural Network has been designed to effectively estimate the states of a nonlinear single input plant. The Unscented Kalman Filter algorithm has been used to train the weights of the Dynamic Neural Network. No a priori bounds of the weights and errors are required. The non linear plant and the Dynamic Neural Network model have been simulated by the same input to illustrate the validity of the theoretical results.

With the ubiquity of complex networks in domains as diverse as social network analysis, recommender systems and epidemiology, improvements in link predictions have a potential for far-reaching impact. In machine learning, a well-known approach to improve classification performance is to combine several methods in an ensemble. It is also well established that, diversity or complementarity of the classifiers to be combined, is a prerequisite for a potential performance improvement of the formed ensemble. In this work, we combine link prediction heuristics (e.g. Common Neighbour, Adamic-Adar etc.), with static graph neural networks, discrete dynamic graph neural networks, and continuous dynamic graph neural networks. We analyze the importance of each of the methods and how complementary they are with respect to each other. We also use greedy searches to select better subsets of methods to combine and train the ensemble in a roll-forward manner to make it adapt to changes in the dynamic network over time. In the conducted experiments on a number of challenging link prediction tasks, the ensemble from the greedy search and the roll-forward ensemble shows on average a 55% and 61% improvement respectively, over the best single method.

Differential inclusions-based dynamic feedback neural network models are introduced to solve in real time nonsmooth convex optimization problems restricted on a closed convex subset of R n . First,a differential inclusion-based dynamic feedback neural network model for solving unconstrained optimization problem is established, and its stability and convergence are investigated, then based on the preceding results and the method of successive approximation, differential inclusions-based dynamic feedback neural network models for solving in real time nonsmooth optimization problem on a closed convex subset are successively constructed, and its dynamical behavior and optimization capabilities are analyzed rigorously.

An increasing number of applications of dynamic neural networks has been developed for digital signal processing (DSP), dynamic neural networks are feedforward neural networks with commonly used scalar synapses replaced by linear filters. This provides feedforward neural networks with the capability of performing dynamic mappings, which depend on past input values, dynamic neural networks are suitable for time series prediction, nonlinear system identification, and signal processing applications. Their most popular types are Finite Impulse Response (FIR) neural networks, which are obtained by replacing synapses with finite impulse response filters. Due to their guaranteed stability characteristic and easy to minimize error surface they have been used with great success in many applications such as signal enhancement, noise cancellation, classification of input patterns, system identification, prediction, and control.. Most of the works on system identification using neural networks are based on multilayer feedforward neural networks with backpropagation learning or more efficient variations of this algorithm, an elegant method for training layered networks. This paper is based on work in a Dynamic System Modeling (DYSMO) and as an application for speed control of DC motor drive.

Dynamic neural network controller model of PEM fuel cell system

Several mathematical approaches to studying analytically the dynamics of neural networks rely on mean-field approximations, which are rigorously applicable only to networks of infinite size. However, all existing real biological networks have finite size, and many of them, such as microscopic circuits in invertebrates, are composed only of a few tens of neurons. Thus, it is important to be able to extend to small-size networks our ability to study analytically neural dynamics. Analytical solutions of the dynamics of small-size neural networks have remained elusive for many decades, because the powerful methods of statistical analysis, such as the central limit theorem and the law of large numbers, do not apply to small networks. In this article, we critically review recent progress on the study of the dynamics of small networks composed of binary neurons. In particular, we review the mathematical techniques we developed for studying the bifurcations of the network dynamics, the dualism between neural activity and membrane potentials, cross-neuron correlations, and pattern storage in stochastic networks. Then, we compare our results with existing mathematical techniques for studying networks composed of a finite number of neurons. Finally, we highlight key challenges that remain open, future directions for further progress, and possible implications of our results for neuroscience.

This paper investigates the use of static and dynamic neural networks in phoneme recognition. Besides this, the paper also proposes a cooperative static and dynamic neural networks model. The cooperative model integrates a decision system for phoneme recognition. Mel cepstrum coding has been applied to represent speech signal in frames. Features from the selected frames are used to train neural networks based models. The comparative study show that the proposed cooperative model provides more accurate recognition rates both in auto-coherence test and generalization test.

We present a methodology to investigate the structure of dynamic networks in terms of concentration of changes in the network. We handle dynamic networks as series of graphs on a fixed set of nodes and consider the changes occurring between two consecutive graphs in the series. We apply our methodology to various dynamic contact networks coming from different contexts and we show that changes in these networks exhibit a non-trivial structure: they are not spread all over the network but are instead concentrated around a small fraction of nodes. We compare our observations on real-world networks to three classical dynamic network models and show that they do not capture this key property.

Information dissemination has become one of the most important services of communication networks. Modeling the diffusion of information through such networks is crucial for our modern information societies. In this work, novel models, segregating between useful and malicious types of information, are introduced, in order to better study Information Dissemination Dynamics (IDD) in wireless complex communication networks, and eventually allow taking into account special network features in IDD. According to the proposed models, and inspired from epidemiology, we investigate the IDD in various complex network types through the use of the Susceptible-Infected (SI) paradigm for useful information dissemination and the Susceptible-Infected-Susceptible (SIS) paradigm for malicious information spreading. We provide analysis and simulation results for both types of diffused information, in order to identify performance and robustness potentials for each dissemination process with respect to the characteristics of the underlying complex networking infrastructures. We demonstrate that the proposed approach can generically characterize IDD in wireless complex networks and reveal salient features of dissemination dynamics in each network type, which could eventually aid in the design of more advanced, robust, and efficient networks and services.

A dynamic pore network model, capable of predicting the displacement of oil from a porous medium by a wettability-altering and interfacial tension reducing surfactant solution, is presented. The key ingredients of the model are (1) a dynamic network model for the displacement of oil by aqueous phase taking account of capillary and viscous effects, (2) a simulation of the transport of surfactant through the network by advection and diffusion taking account of adsorption on the solid surface, and (3) the coupling of these two by linking the contact angle and interfacial tension appearing in the dynamic network simulation to the local concentration of surfactant computed in the transport simulation. The coupling is two-way: The flow field used to advect the surfactant concentration is that associated with the displacement of oil by the injected aqueous phase, and the surfactant concentration influences the flow field through its effect on the capillarity parameters. We present results obtained using the model to validate that it reproduces the displacement patterns observed by other authors in two-dimensional networks as capillary number and mobility ratio are varied, and to illustrate the effects of surfactant on displacement patterns. A mechanism is demonstrated whereby in an initially mixed-wet medium, surfactant-induced wettability alteration can lead to stabilization of displacement fronts.

Despite increased focus on well-being in the psychological sciences in recent years, cross-sectional and group-level methods fall short of capturing the dynamics of individual people’s well-being in daily life. Such an investigation is especially important for young adults who are in a developmental phase in which they are more likely to experience fluctuations in their daily lives. We used ecological momentary assessment over one week in two samples of first-year college students (Sample 1, N = 103, assessments = 2,535; Sample 2, N = 76, assessments = 1,796) and a dynamic network approach (Dynamic Exploratory Graph Analysis). This approach estimates, within each person, how well-being elements fluctuate across days and co-vary with one another, recovering both their differentiation (exclusivity) and their interdependence (connectivity). In this approach, elements of well-being—measured via the PERMA framework—are modeled as a dynamic network to address the following research questions: (1) How do elements of well-being in young adults’ daily life form a dynamic network? (2) Does the structure of well-being hold across all young adults, or is there heterogeneity? (3) Is there synchrony among well-being elements, and do they drive a person’s well-being network? Findings suggest that while group-level dynamic well-being networks align with theoretical models, each person’s individual experiences vary significantly, with each person demonstrating unique well-being network structures and synchronous elements. These findings underscore the importance of considering individual variability in well-being, highlighting the necessity for personalized approaches to understand and promote well-being among young adults.

Genetic algorithms are used to search for network weights which cause the dynamical network to produce long attractors. Several variations of the genetic algorithm are described, and the search performance is compared to that of the base-line method of randomly selected weights. It is pointed out that dynamical networks support self-sustaining patterns of oscillation which can be initiated by a one-time input strobe. These self-sustaining patterns, or attractor cycles, evolve into a repeating pattern for most combinations of network weights and input strobes. Attractor cycles vary in length and are a function of the particular network weights and the particular strobe. An interesting property of these networks is that a particular set of network weights can produce, or recall, a variety of repeating patterns, where the one that is evoked depends on the triggering strobe. This effectively is the storage of sequential patterns in the form of attractors. >

One of the difficulties encountered in the application of reinforcement learning methods to real-world problem is the generalization of large-scale or continuous state space. In order to solve the curse of dimensionality problem caused by discretizing continuous state space, a kind of Q-learning method for continuous state space based on a dynamic Elman neural network was proposed in this paper. The inputs and the output of Elman network are the system state-action pair and the corresponding Q-value. That is, Elman network is used to estimate the Q-value of state-action pair on-line. Eligibility trace for connecting weights is introduced by borrowing ideas from the eligibility trace mechanism of state in temporal difference algorithm to improve the learning speed of neural network. Computer simulations on mountain car control illustrate the performance and applicability of the proposed Q-learning scheme.