Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session Q09: Predicting Nonlinear and Complex Systems with Machine Learning IIIFocus Recordings Available

Hide Abstracts 
Sponsoring Units: GSNP DSOFT DCOMP Chair: YingCheng Lai, Arizona State University Room: McCormick Place W180 
Wednesday, March 16, 2022 3:00PM  3:36PM 
Q09.00001: Nextgeneration reservoir computing Invited Speaker: Daniel J Gauthier Reservoir computing is a bestinclass machine learning algorithm for processing information generated by dynamical systems using observed timeseries data. Importantly, it requires very small training data sets, uses linear optimization, and thus requires minimal computing resources. However, the algorithm uses randomly sampled matrices to define the underlying recurrent neural network and has a multitude of metaparameters that must be optimized. Recent results demonstrate the equivalence of reservoir computing to nonlinear vector autoregression, which requires no random matrices, fewer metaparameters, and provides interpretable results. Here, we demonstrate that nonlinear vector autoregression excels at reservoir computing benchmark tasks and requires even shorter training data sets and training time, heralding the next generation of reservoir computing. 
Wednesday, March 16, 2022 3:36PM  3:48PM 
Q09.00002: Divergent Predictive States: The Statistical Complexity Dimension of Stationary, Ergodic Hidden Markov Processes Alexandra M Jurgens, James P Crutchfield Even simplydefined, finitestate generators produce stochastic processes that 
Wednesday, March 16, 2022 3:48PM  4:00PM 
Q09.00003: Infer dynamics in experimental dusty plasma trajectories by machine learning Wentao Yu, Justin C Burton Advanced machine learning models have been developed to infer governing equations from time series, but they are mostly tested on simulated data since experimental data are not labeled. Additionally, experimental data may involve drift and nonGaussian noise, which is difficult to simulate. To bridge this divide, we benchmark the prediction of simple machine learning models using the stochastic motion of micronsized dust particles levitated in a plasma. The particles often experience conservative and nonconservative complex forces which are not wellunderstood. Importantly, we label the data using an alternative experimental method involving mechanical perturbation. In experiments with a single particle, machine learning predicts the forces as precisely as in simulated data. Experiments with many particles reveal the interaction force, which is related to the particle charge and plasma Debye length, and are difficult to measure insitu. Nevertheless, our results show that machine learning can accurately predict the system parameters when the digitallyimaged random motion of the particles is less than 2 pixels. 
Wednesday, March 16, 2022 4:00PM  4:12PM 
Q09.00004: Mechanical memory manipulation using Reinforcement Learning Laura Michel, Frédéric Lechenault, Théo Jules

Wednesday, March 16, 2022 4:12PM  4:24PM 
Q09.00005: Noise, data and time resolution limits for reconstruction of allatom protein structures from single molecule FRET time series Maximilian T Topel, Andrew L Ferguson Single molecule FRET (smFRET) is an experimental technique used to record protein dynamics by periodically measuring a small set of scalar observables. Takens' Theorem states that under specific conditions, time series vectors constructed from scalar observables of a dynamical system embed full dimensional system dynamics. Single molecule Takens' reconstruction (STAR) combines smFRET measurements with Takens' Theorem to reconstruct the all atom molecular trajectory through manifold learning and artificial neural networks. Applying synthetic noise characteristic of smFRET, time binned observation data, and limited learning data, we employ molecular dynamics simulations to test the experimental limits of STAR in terms of data volume, time resolution, and signaltonoise ratio to place limits on anticipated reconstruction accuracy. Our results show that Chignolin can be reconstructed with RMSDs of 0.2 nm, Villin to within 0.4 nm, and BBL to within 0.5 nm under conditions representative of realistic smFRET experiments. 
Wednesday, March 16, 2022 4:24PM  4:36PM 
Q09.00006: Novel Deep Learning approaches for Complex Random Telegraph Noise Analysis Lu Wang, Marcel J Robitaille, HeeBong Yang, Na Young Kim Machine learning, especially deep learning, have been rapidly developed in recent ten years in various aspects, from computer vision to natural language processing, and offers a powerful tool for us to solve challenging tasks in wide applications. Upon this successful demonstration, we design a sequence analysis structure based on deep learning technology for efficient analysis protocol to investigate complex random telegraph noise signals (RTN). RTNs appear prevalent in many classical and quantum devices. In a traditional method, it is a big challenge to extract quantitative information of each trap from the RTN signals in the presence of white noise and to detect transition rates accurately for multiple traps. Here we overcome this challenge by building a sequence analysis model using a Wavenet structure, and we extract signal amplitudes and time constants of many trap signals with multiple states with high accuracies. 
Wednesday, March 16, 2022 4:36PM  4:48PM 
Q09.00007: Modeling a Physical Chaotic System by Measuring its Dynamics in Time using Gradient Descent Algorithms Roie Ezraty, Shmuel M Rubinstein A nonlinear electrical circuit consisting of a resistor, inductor and diode exhibits complicated dynamics such as period doubling bifurcations and chaos. Due to the nonlinearity of the circuit components, especially the diode, such a circuit is hard to model physically. Nonetheless, using gradient descent algorithms on voltage measurements over the different components of the circuit, we solve the inverse problem and give a dynamical model of the circuit. The model is in the form of an ordinary differential equation dependant only on standard variables in electric circuits such as voltage and cummulative charge on the circuit components. Our method can be further implemented to other systems that show intrinsically unpredictable dynamics. 
Wednesday, March 16, 2022 4:48PM  5:00PM 
Q09.00008: Estimating covariant Lyapunov vectors from data Nahal S Sharafi, Christoph Martin Covariant Lyapunov vectors characterize the directions along which perturbations in dynamical systems grow. They have also been studied as predictors of critical transitions and extreme events. For many applications like, for example, prediction, it is necessary to estimate the vectors from data since model equations are unknown for many interesting phenomena. We propose a novel method for estimating covariant Lyapunov vectors based on data records without knowing the underlying equations of the system. In contrast to previous approaches, our approach can be applied to highdimensional datasets. We demonstrate that this purely datadriven approach can accurately estimate covariant Lyapunpov vectors from data records generated by low and highdimensional dynamical systems. The highest dimension of a timeseries from which covariant Lyapunov vectors were estimated in this contribution is 128. Being able to infer covariant Lyapunov vectors from datarecords could encourage numerous future applications in dataanalysis and databased predictions. 
Wednesday, March 16, 2022 5:00PM  5:12PM 
Q09.00009: Mutual information disentangles internal interactions from changing environments Daniel Maria Busiello, Giorgio Nicoletti Realworld systems are characterized by complex interactions among their internal degrees of freedom while living in everchanging environments whose net effect is to act as additional couplings. Here, we study a paradigmatic interacting model in a switching, but unobserved, environment. We show that the limiting properties of the mutual information of the system allow for a disentangling of these two sources of couplings. Our approach can be extended to a wide class of stochastic dynamics and might stand as a general method to discriminate complex internal interactions from equally complex changing environments. 
Wednesday, March 16, 2022 5:12PM  5:24PM 
Q09.00010: Topological Boundary Constrais in Artificial Colloidal Ice Carolina RodríguezGallo, Pietro Tierno, Antonio OrtizAmbriz Recently, the research of collective phenomena has begun to exploit the new advances in micro and nanoscience to tailor and design systems with desired properties, often more fertile than their “natural” counterparts. I will show our latest work on an Artificial Colloidal Ice system, where, using Brownian Dynamic simulations and “proof of concept” experiments, we study the effect of boundaries in a geometrically frustrated system and how they can be used to influence the bulk behavior. 
Follow Us 
Engage
Become an APS Member 
My APS
Renew Membership 
Information for 
About APSThe American Physical Society (APS) is a nonprofit membership organization working to advance the knowledge of physics. 
© 2023 American Physical Society
 All rights reserved  Terms of Use
 Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 207403844
(301) 2093200
Editorial Office
1 Research Road, Ridge, NY 119612701
(631) 5914000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 200452001
(202) 6628700