arXiv subjects

Jeff Moehlis

Publications and source records attributed to Jeff Moehlis.

14 recordsLinked to original sources

arXiv2026

Symbolic Neural ODEs: Learning interpretable models from time-series data

We present a machine learning framework for identifying sparse, interpretable models of dynamical systems directly from time-series data. Our approach parameterizes the underlying vector field using a neural architecture and trains it by minimizing a multi-step prediction loss over a finite horizon. To ensure numerical tractability, we optimize a mean absolute error objective averaged across prediction steps, and progressively increase the horizon during training. A key feature of this formulation is that it enforces consistency under repeated composition of the learned dynamics. As a result, the identified models exhibit significantly improved stability compared with approaches based on one-step regression of the vector field. When combined with sparsity-promoting regularization, this leads to parsimonious models that generalize beyond the training data. We demonstrate accurate recovery of systems exhibiting a wide range of behaviors, including stable and unstable fixed points, periodic orbits, and chaotic attractors. For chaotic systems, while long-term trajectory prediction is inherently limited by sensitivity to initial conditions, we show that multi-step training yields models with accurate short-term dynamics and strong agreement in long-time statistical properties, including mean, variance, and Lyapunov exponents. Moreover, we establish theoretical bounds linking trajectory error to statistical accuracy, providing a step toward a principled explanation for this behavior.

cs.LG↗

arXiv2026

Symbolic Regression via Neural Networks

Identifying governing equations for a dynamical system is a topic of critical interest across an array of disciplines, from mathematics to engineering to biology. Machine learning -- specifically deep learning -- techniques have shown their capabilities in approximating dynamics from data, but a shortcoming of traditional deep learning is that there is little insight into the underlying mapping beyond its numerical output for a given input. This limits their utility in analysis beyond simple prediction. Simultaneously, a number of strategies exist which identify models based on a fixed dictionary of basis functions, but most either require some intuition or insight about the system, or are susceptible to overfitting or a lack of parsimony. Here we present a novel approach that combines the flexibility and accuracy of deep learning approaches with the utility of symbolic solutions: a deep neural network that generates a symbolic expression for the governing equations. We first describe the architecture for our model, then show the accuracy of our algorithm across a range of classical dynamical systems.

math.DS↗

arXiv2025

Nearly Optimal Chaotic Desynchronization of Neural Oscillators

Motivated by deep brain stimulation treatment of neural disorders such as Parkinson's disease, it has been proposed that desynchronization of neural oscillators can be achieved by maximizing the Lyapunov exponent of the phase difference between pairs of oscillators. Here we consider two approximations to optimal stimuli for chaotic desynchronization of neural oscillators. These approximations are based on the oscillators' phase response curve, and unlike previous approaches do not require numerical solution of a two-point boundary value problem. It is shown that these approximations can achieve nearly optimal desynchronization, and can be used with an event-based control scheme to desynchronize populations of noisy, coupled neurons.

math.OC↗

arXiv2025

CASL-HJX: A Comprehensive Guide to Solving Deterministic and Stochastic Hamilton-Jacobi Equations

CASL-HJX is a computational framework designed for solving deterministic and stochastic Hamilton-Jacobi equations in two spatial dimensions. It provides a flexible and efficient approach to modeling front propagation problems, optimal control problems, and stochastic Hamilton-Jacobi Bellman equations. The framework integrates numerical methods for hyperbolic PDEs with operator splitting techniques and implements implicit methods for second-order derivative terms, ensuring convergence to viscosity solutions while achieving global rather than local optimization. Built with a high-performance C++ core, CASL-HJX efficiently handles mixed-order derivative systems with time-varying dynamics, making it suitable for real-world applications across multiple domains. We demonstrate the solver's versatility through tutorial examples covering various PDEs and through applications in neuroscience, where it enables the design of energy-efficient controllers for regulating neural populations to mitigate pathological synchrony. While our examples focus on these applications, the mathematical foundation of the solver makes it applicable to problems in finance, engineering, and machine learning. The modular architecture allows researchers to define computational domains, configure problems, and execute simulations with high numerical accuracy. CASL-HJX bridges the gap between deterministic control methods and stochastic models, providing a robust tool for managing uncertainty in complex dynamical systems.

math.OC↗

arXiv2024

Expressive Symbolic Regression for Interpretable Models of Discrete-Time Dynamical Systems

Interpretable mathematical expressions defining discrete-time dynamical systems (iterated maps) can model many phenomena of scientific interest, enabling a deeper understanding of system behaviors. Since formulating governing expressions from first principles can be difficult, it is of particular interest to identify expressions for iterated maps given only their data streams. In this work, we consider a modified Symbolic Artificial Neural Network-Trained Expressions (SymANNTEx) architecture for this task, an architecture more expressive than others in the literature. We make a modification to the model pipeline to optimize the regression, then characterize the behavior of the adjusted model in identifying several classical chaotic maps. With the goal of parsimony, sparsity-inducing weight regularization and information theory-informed simplification are implemented. We show that our modified SymANNTEx model properly identifies single-state maps and achieves moderate success in approximating a dual-state attractor. These performances offer significant promise for data-driven scientific discovery and interpretation.

cs.LG↗

arXiv2020

Analysis of Neural Clusters due to Deep Brain Stimulation Pulses

Deep brain stimulation (DBS) is an established method for treating pathological conditions such as Parkinson's disease, dystonia, Tourette syndrome, and essential tremor. While the precise mechanisms which underly the effectiveness of DBS are not fully understood, theoretical studies of populations of neural oscillators stimulated by periodic pulses suggest that this may be related to clustering, in which subpopulations of the neurons are synchronized, but the subpopulations are desynchronized with respect to each other. The details of the clustering behavior depend on the frequency and amplitude of the stimulation in a complicated way. In the present study, we investigate how the number of clusters, their stability properties, and their basins of attraction can be understood in terms of one-dimensional maps defined on the circle. Moreover, we generalize this analysis to stimuli that consist of pulses with alternating properties, which provide additional degrees of freedom in the design of DBS stimuli. Our results illustrate how the complicated properties of clustering behavior for periodically forced neural oscillator populations can be understood in terms of a much simpler dynamical system.

math.DS↗

arXiv2020

Supervised Learning Algorithms for Controlling Underactuated Dynamical Systems

Control of underactuated dynamical systems has been studied for decades in robotics, and is now emerging in other fields such as neuroscience. Most of the advances have been in model based control theory, which has limitations when the system under study is very complex and it is not possible to construct a model. This calls for data driven control methods like machine learning, which has spread to many fields in the recent years including control theory. However, the success of such algorithms has been dependent on availability of large datasets. Moreover, due to their black box nature, it is challenging to analyze how such algorithms work, which may be crucial in applications where failure is very costly. In this paper, we develop two related novel supervised learning algorithms. The algorithms are powerful enough to control a wide variety of complex underactuated dynamical systems, and yet have a simple and intelligent structure that allows them to work with a sparse data set even in the presence of noise. Our algorithms output a bang-bang (binary) control input by taking in feedback of the state of the dynamical system. The algorithms learn this control input by maximizing a reward function in both short and long time horizons. We demonstrate the versatility of our algorithms by applying them to a diverse range of applications including: switching between bistable states, changing the phase of an oscillator, desynchronizing a population of synchronized coupled oscillators, and stabilizing an unstable fixed point. For most of these applications we are able to reason why our algorithms work by using traditional dynamical systems and control theory. We also compare our learning algorithms with some traditional control algorithms, and reason why our algorithms work better.

math.OC↗

arXiv2020

Augmented Phase Reduction for Periodic Orbits Near a Homoclinic Bifurcation and for Relaxation Oscillators

Oscillators - dynamical systems with stable periodic orbits - arise in many systems of physical, technological, and biological interest. The standard phase reduction, a model reduction technique based on isochrons, can be unsuitable for oscillators which have a small-magnitude negative nontrivial Floquet exponent. This necessitates the use of the augmented phase reduction, a recently devised two-dimensional reduction technique based on isochrons and isostables. In this article, we calculate analytical expressions for the augmented phase reduction for two dynamically different planar systems: periodic orbits born out of a homoclinic bifurcation, and relaxation oscillators. To validate our calculations, we simulate models in these dynamic regimes, and compare their numerically computed augmented phase reduction with the derived analytical expressions. These analytical and numerical calculations help us to understand conditions for which the use of augmented phase reduction over the standard phase reduction can be advantageous.

math.DS↗

arXiv2019

Phase Distribution Control of a Population of Oscillators

The collective behavior of biological oscillators has been recognized as an important problem for several decades, but its control has come into limelight only recently. Much of the focus for control has been on desynchronization of an oscillator population, motivated by the pathological neural synchrony present in essential and parkinsonian tremor. Other applications, such as the beating of the heart and insulin secretion, require synchronization, and recently there has been interest in forming clusters within an oscillator population as well. In this article, we use a formulation that allows us to devise control frameworks to achieve all of these distinct collective behaviors observed in biological oscillators. This is based on the Fourier decomposition of the partial differential equation governing the evolution of the phase distribution of a population of identical, uncoupled oscillators. Our first two control algorithms are Lyapunov-based, which work by decreasing a positive definite Lyapunov function towards zero. Our third control is an optimal control algorithm, which minimizes the control energy consumption while achieving the desired collective behavior of an oscillator population. Motivated by pathological neural synchrony, we apply our control to desynchronize an initially synchronized neural population. Given the proposed importance of enhancing spike time dependent plasticity to stabilize neural clusters and counteract pathological neural synchronization, we formulate the phase difference distribution in terms of the phase distribution, and prove some of its fundamental properties, and in turn apply our control to transform the neural phase distribution to form clusters. Finally, motivated by eliminating cardiac alternans, we apply our control to phase shift a synchronous cardiac pacemaker cell population.

nlin.AO↗

arXiv2014

A Nonparametric Adaptive Nonlinear Statistical Filter

We use statistical learning methods to construct an adaptive state estimator for nonlinear stochastic systems. Optimal state estimation, in the form of a Kalman filter, requires knowledge of the system's process and measurement uncertainty. We propose that these uncertainties can be estimated from (conditioned on) past observed data, and without making any assumptions of the system's prior distribution. The system's prior distribution at each time step is constructed from an ensemble of least-squares estimates on sub-sampled sets of the data via jackknife sampling. As new data is acquired, the state estimates, process uncertainty, and measurement uncertainty are updated accordingly, as described in this manuscript.

stat.ML↗

arXiv2013

Phase noise of oscillators with unsaturated amplifiers

We study the role of amplifier saturation in eliminating feedback noise in self-sustained oscillators. We extend previous works that use a saturated amplifier to quench fluctuations in the feedback magnitude, while simultaneously tuning the oscillator to an operational point at which the resonator nonlinearity cancels fluctuations in the feedback phase. We consider a generalized model which features an amplitude-dependent amplifier gain function. This allows us to determine the total oscillator phase noise in realistic configurations due to noise in both quadratures of the feedback, and to show that it is not necessary to drive the resonator to large oscillation amplitudes in order to eliminate noise in the phase of the feedback.

cond-mat.other↗

arXiv2013

Isostables, isochrons, and Koopman spectrum for the action-angle representation of stable fixed point dynamics

For asymptotically periodic systems, a powerful (phase) reduction of the dynamics is obtained by computing the so-called isochrons, i.e. the sets of points that converge toward the same trajectory on the limit cycle. Motivated by the analysis of excitable systems, a similar reduction has been attempted for non-periodic systems admitting a stable fixed point. In this case, the isochrons can still be defined but they do not capture the asymptotic behavior of the trajectories. Instead, the sets of interest-that we call isostables-are defined in literature as the sets of points that converge toward the same trajectory on the stable slow manifold of the fixed point. However, it turns out that this definition of the isostables holds only for systems with slow-fast dynamics. Also, efficient methods for computing the isostables are missing. The present paper provides a general framework for the definition and the computation of the isostables of stable fixed points, which is based on the spectral properties of the so-called Koopman operator. More precisely, the isostables are defined as the level sets of a particular eigenfunction of the Koopman operator. Through this approach, the isostables are unique and well-defined objects related to the asymptotic properties of the system. Also, the framework reveals that the isostables and the isochrons are two different but complementary notions which define a set of action-angle coordinates for the dynamics. In addition, an efficient algorithm for computing the isostables is obtained, which relies on the evaluation of Laplace averages along the trajectories. The method is illustrated with the excitable FitzHugh-Nagumo model and with the Lorenz model. Finally, we discuss how these methods based on the Koopman operator framework relate to the global linearization of the system and to the derivation of special Lyapunov functions.

math.DS↗

arXiv2001

Effect of Noise on Excursions To and Back From Infinity

The effect of additive white noise on a model for bursting behavior in large aspect-ratio binary fluid convection is considered. Such bursts are present in systems with nearly square symmetry and are the result of heteroclinic cycles involving infinite amplitude states created when the square symmetry is broken. A combination of numerical results and analytical arguments show how even a very small amount of noise can have a very large effect on the amplitudes of successive bursts. Large enough noise can also affect the physical manifestations of the bursts. Finally, it is shown that related bursts may occur when white noise is added to the normal form equations for the Hopf bifurcation with exact square symmetry.

nlin.CD↗

arXiv1999

Radiation of mixed layer near-inertial oscillations into the ocean interior

The radiation from the mixed layer into the interior of the ocean of near-inertial oscillations excited by a passing storm in the presence of the beta effect is reconsidered as an initial-value problem. Making use of the fact that the mixed layer depth is much smaller than the total depth of the ocean, the solution is obtained in the limit of an ocean that is effectively infinitely deep. For a uniform initial condition, analytical results for the velocity, horizontal kinetic energy density and fluxes are obtained. The resulting decay of near-inertial mixed layer energy in the presence of the beta effect occurs on a timescale similar to that observed.

physics.ao-ph↗