Nonlinear tracking control enabling a dynamical system to track desired trajectory is fundamental robotics, serving wide range of civil and defense applications. In engineering, designing requires complete knowledge the model equations. We develop model-free, machine-learning framework two-arm robotic manipulator using only partially observed states, where controller realized by reservoir computing. Stochastic input exploited for training, which consists partial state vector as first its...

The benefits of noise to applications nonlinear dynamical systems through mechanisms such as stochastic and coherence resonances have been well documented. Recent years witnessed a growth research in exploiting machine learning predict systems. It has known that can act regularizer improve the training performance learning. Utilizing reservoir computing paradigm, we find injecting data induce resonance phenomenon with significant both short-term prediction state variables long-term...

Detecting a weak physical signal immersed in overwhelming noises entails separating the two, task for which machine learning is naturally suited. In principle, such generated by nonlinear dynamical system of intrinsically high dimension mathematical model not available, rendering unsuitable traditional linear or state-estimation methods that require an accurate (e.g., extended Kalman filters). We exploit architectures reservoir computing and feed-forward neural networks (FNNs) with...

It has been recently demonstrated that two machine-learning architectures, reservoir computing and time-delayed feed-forward neural networks, can be exploited for detecting the Earth’s anomaly magnetic field immersed in overwhelming complex signals navigation a GPS-denied environment. The accuracy of detected corresponds to positioning range 10–40 m. To increase reduce uncertainty weak signal detection as well directly obtain position information, we exploit model random forests combines...

Complex and nonlinear dynamical systems often involve parameters that change with time, accurate tracking of which is essential to tasks such as state estimation, prediction, control. Existing machine-learning methods require full observation the underlying system tacitly assume adiabatic changes in parameter. Formulating an inverse problem exploiting reservoir computing, we develop a model-free fully data-driven framework accurately track time-varying from partial real time. In particular,...

Reinforcement learning (RL) has been employed to devise the best course of actions in defending critical infrastructures, such as power networks against cyberattacks. Nonetheless, even case smallest grids, action space RL experiences exponential growth, rendering efficient exploration by agent practically unattainable. The current algorithms tailored grids are generally not suited when state-action size becomes large, despite trade-offs. We address large action-space problem for grid...

Anticipating a tipping point, transition from one stable steady state to another, is problem of broad relevance due the ubiquity phenomenon in diverse fields. The steady-state nature dynamics about point makes its prediction significantly more challenging than predicting other types critical transitions oscillatory or chaotic dynamics. Exploiting benefits noise, we develop general data-driven and machine-learning approach potential future nonautonomous dynamical systems validate framework...

<title>Abstract</title> In applications, an anticipated issue is where the system of interest has never been encountered before and sparse observations can be made only once. Can dynamics faithfully reconstructed? We address this challenge by developing a hybrid transformer reservoir-computing scheme. The trained without using data from target system, but with essentially unlimited synthetic known chaotic systems. then tested its output further fed into reservoir computer for predicting...

The ever-increasing complexity of modern power grids makes them vulnerable to cyber and/or physical attacks. To protect them, accurate attack detection is essential. A challenging scenario that a localized has occurred on specific transmission line but only small number lines elsewhere can be monitored. That is, full state observation the whole grid not feasible, so and estimation need done with limited, partial observations. We articulate machine-learning framework address this problem,...

According to the official report, first case of COVID-19 and death in United States occurred on January 20 February 29, 2020, respectively. On April 21, California reported that state 6, implying community spreading might have started earlier than previously thought. Exactly what is time zero, i.e., when did emerge begin spread U.S. other countries? We develop a comprehensive predictive modeling framework address this question. Using available data confirmed infections obtain optimal values...

An urgent problem in controlling COVID-19 spreading is to understand the role of undocumented infection. We develop a five-state model for COVID-19, taking into account unique features novel coronavirus, with key parameters determined by government reports and mathematical optimization. Tests using data from China, South Korea, Italy, Iran indicate that capable generating accurate prediction daily accumulated number confirmed cases entirely suitable real-time prediction. The drastically...

Recent research on the Atlantic Meridional Overturning Circulation (AMOC) raised concern about its potential collapse through a tipping point due to climate-change caused increase in freshwater input into North Atlantic. The predicted time window of is centered middle century and earliest possible start approximately two years from now. More generally, anticipating at which system transitions one stable steady state another relevant broad range fields. We develop machine-learning approach...

The Atlantic Meridional Overturning Circulation (AMOC) is a significant component of the global ocean system, which has so far ensured relatively warm climate for North and mild conditions in regions, such as Western Europe. AMOC also critical climate. complexity dynamical system underlying vast that long-term assessment potential risk collapse extremely challenging. However, short-term prediction can lead to accurate estimates state possibly early warning signals guiding policy making...

We uncover a phenomenon in coupled nonlinear networks with symmetry: as bifurcation parameter changes through critical value, synchronization among subset of nodes can deteriorate abruptly, and, simultaneously, perfect emerges suddenly different that are not directly connected. This is metamorphosis leading to an explosive transition remote synchronization. The finding demonstrates onset synchrony and synchronization, two phenomena have been studied separately, arise the same system due...

Symmetries, due to their fundamental importance dynamical processes on networks, have attracted a great deal of current research. Finding all symmetric nodes in large complex networks typically relies automorphism groups from algebraic-group theory, which are solvable quasipolynomial time. We articulate conceptually appealing and computationally extremely efficient approach finding characterizing by introducing structural position vector (SPV) for each node networks. establish the...

It was recently demonstrated that two machine-learning architectures, reservoir computing and time-delayed feed-forward neural networks, can be exploited for detecting the Earth's anomaly magnetic field immersed in overwhelming complex signals navigation a GPS-denied environment. The accuracy of detected corresponds to positioning range 10 40 meters. To increase reduce uncertainty weak signal detection as well directly obtain position information, we exploit model random forests combines...

A foundational machine-learning architecture is reinforcement learning, where an outstanding problem achieving optimal balance between exploration and exploitation. Specifically, enables the agents to discover policies in unknown domains of environment for gaining potentially large future rewards, while exploitation relies on already acquired knowledge maximize immediate rewards. We articulate approach this problem, treating dynamical process learning as a Markov decision that can be modeled...

A fundamental challenge in developing data-driven approaches to ecological systems for tasks such as state estimation and prediction is the paucity of observational or measurement data. For example, modern machine-learning techniques deep learning reservoir computing typically require a large quantity Leveraging synthetic data from paradigmatic nonlinear but non-ecological dynamical systems, we develop meta-learning framework with time-delayed feedforward neural networks predict long-term...

In applications, an anticipated situation is where the system of interest has never been encountered before and sparse observations can be made only once. Can dynamics faithfully reconstructed from limited without any training data? This problem defies known traditional methods nonlinear time-series analysis as well existing machine-learning that typically require extensive data target for training. We address this challenge by developing a hybrid transformer reservoir-computing scheme. The...

Abstract Due to the heterogeneity among States in US, predicting COVID-19 trends and quantitatively assessing effects of government testing capability control measures need be done via a State-by-State approach. We develop comprehensive model for incorporating time delays population movements. With key parameter values determined by empirical data, enables most likely epidemic scenarios predicted each State, which are indicative whether services vigorous enough contain disease. find that...

Abstract According to the official report, first case of COVID-19 and death in United States occurred on January 20 February 29, 2020, respectively. On April 21, California reported that state 6, implying community spreading might have started earlier than previously thought. Exactly what is time ZERO, i.e., when did emerge begin spread US other countries? We develop a comprehensive predictive modeling framework address this question. Using available data confirmed infections obtain optimal...

Can noise be beneficial to machine-learning prediction of chaotic systems? Utilizing reservoir computers as a paradigm, we find that injecting the training data can induce stochastic resonance with significant benefits both short-term state variables and long-term attractor system. A key inducing is include amplitude in set hyperparameters for optimization. By so doing, accuracy, stability horizon dramatically improved. The phenomenon demonstrated using two prototypical high-dimensional systems.

Symmetries are fundamental to dynamical processes in complex networks such as cluster synchronization, which have attracted a great deal of current research. Finding symmetric nodes large networks, however, has relied on automorphism groups algebraic group theory, solvable quasipolynomial time. We articulate conceptually appealing and computationally extremely efficient approach finding characterizing all by introducing structural position vector (SPV) for each every node the network. prove...

Abstract Nonlinear tracking control enabling a dynamical system to track desired trajectory is fundamental robotics, serving wide range of civil and defense applications. In engineering, designing requires complete knowledge the model equations. We develop model-free, machine-learning framework two-arm robotic manipulator using only partially observed states, where controller realized by reservoir computing. Stochastic input exploited for training, which consists partial state vector as...