Search NASA⌕ Search

SEARCH · Search NASA

Results for “learning based systems”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,081 records · Page 60

Precision Polishing of Ablator Capsules via in situ Process Monitoring and Machine Learning–Based Optimization

In inertial confinement fusion (ICF) experiments seeking output gains of unity and beyond, the quality of the ablator capsule is paramount for minimizing the hydrodynamic mix that quenches the central hot spot. Defects in the form of foreign particles or missing mass on the surface and within the wall of the capsule are primary offenders. High-density carbon capsules made for ICF experiments at the National Ignition Facility are precision polished to achieve surface smoothness on the order of a few nanometers as well as to minimize isolated defects in the form of pits. Given the critical role of this process, we are developing smart manufacturing techniques with the goal of elevating the efficiency of this process. Our approach is to use MEMS (micro-electromechanical systems)–based sensors to capture the fine vibration signals generated during the polishing process and combine them with synchronized visual feedback as needed. Beyond using these sensors for process monitoring, we use specific deep learning methods to analyze the data and extract correlations with both the process parameters and the final performance of the polishing run. Here, in this work, we describe the multiple fronts we have explored in this regard and the results we have gotten so far. This approach promises to have the potential to ultimately provide real-time feedback that can be used to ensure the progress of the run as well as a means for faster optimization.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Physics-informed machine learning exploration of Na storage mechanisms in disordered carbon

Sodium-ion batteries are a cost-effective, sustainable alternative to lithium-ion systems for large-scale energy storage. However, optimizing sodium storage in carbon-based anodes with microstructural complexity and atomic disorder remains a major challenge. The intrinsic inhomogeneity of these materials produces diverse local environments, making it difficult for conventional methods to predict and control ion dynamics. Hard carbon (HC) anodes, composed of ranges of ordered-to-disordered graphitic and amorphous nanodomains, offer tunable ion storage and rate capacity, yet rationale design remains a challenge due to poorly understood correlation between local atomic feature and ion transport mechanism. Here, to address this challenge, we introduce a data-driven framework that integrates validated machine-learned interatomic potentials, large-scale molecular dynamics simulations, and machine learning to elucidate sodium transport mechanisms as a function of carbon and sodium loading densities. By computing per-ion structural descriptors and applying unsupervised learning, we identify distinct diffusion modes governed by microscopic features. Supervised analysis and correlation mapping then establish quantitative links between these transport regimes and processing variables such as bulk carbon density and sodium content. This physics-informed approach establishes quantitative structure–transport relationships and offers actionable design principles for engineering high-performance HC anodes.

Data-driven framework↗

Data-driven agent-based modeling, with application to rooftop solar adoption

Agent-based modeling is commonly used for studying complex system properties emergent from interactions among many agents. We present a novel data-driven agent-based modeling framework applied to forecasting individual and aggregate residential rooftop solar adoption in San Diego county. Our first step is to learn a model of individual agent behavior from combined data of individual adoption characteristics and property assessment. We then construct an agent-based simulation with the learned model embedded in artificial agents, and proceed to validate it using a holdout sequence of collective adoption decisions. We demonstrate that the resulting agent-based model successfully forecasts solar adoption trends and provides a meaningful quantification of uncertainty about its predictions. We utilize our model to optimize two classes of policies aimed at spurring solar adoption: one that subsidizes the cost of adoption, and another that gives away free systems to low-income house- holds. We find that the optimal policies derived for the latter class are significantly more efficacious, whereas the policies similar to the current California Solar Initiative incentive scheme appear to have a limited impact on overall adoption trends.

Agent-based modeling↗

Bayesian analysis of energy and count rate data for detection of low count rate radioactive sources

We propose a radiation detection system which generates its own discrete sampling distribution based on past measurements of background. The advantage to this approach is that it can take into account variations in background with respect to time, location, energy spectra, detector-specific characteristics (i.e. different efficiencies at different count rates and energies), etc. This would therefore be a 'machine learning' approach, in which the algorithm updates and improves its characterization of background over time. The system would have a 'learning mode,' in which it measures and analyzes background count rates, and a 'detection mode,' in which it compares measurements from an unknown source against its unique background distribution. By characterizing and accounting for variations in the background, general purpose radiation detectors can be improved with little or no increase in cost. The statistical and computational techniques to perform this kind of analysis have already been developed. The necessary signal analysis can be accomplished using existing Bayesian algorithms which account for multiple channels, multiple detectors, and multiple time intervals. Furthermore, Bayesian machine-learning techniques have already been developed which, with trivial modifications, can generate appropriate decision thresholds based on the comparison of new measurements against a nonparametric sampling distribution. (authors)

97 MATHEMATICS AND COMPUTING↗

A survey of unsupervised learning methods for high-dimensional uncertainty quantification in black-box-type problems

Constructing surrogate models for uncertainty quantification (UQ) on complex partial differential equations (PDEs) having inherently high-dimensional O(10 n ), n ≥ 2, stochastic inputs (e.g., forcing terms, boundary conditions, initial conditions) poses tremendous challenges. The “curse of dimensionality” can be addressed with suitable unsupervised learning techniques used as a pre-processing tool to encode inputs onto lower-dimensional subspaces while retaining its structural information and meaningful properties. In this work, we review and investigate thirteen dimension reduction methods including linear and nonlinear, spectral, blind source separation, convex and non-convex methods and utilize the resulting embeddings to construct a mapping to quantities of interest via polynomial chaos expansions (PCE). Here, we refer to the general proposed approach as manifold PCE (m-PCE), where manifold corresponds to the latent space resulting from any of the studied dimension reduction methods. To investigate the capabilities and limitations of these methods we conduct numerical tests for three physics-based systems (treated as black-boxes) having high-dimensional stochastic inputs of varying complexity modeled as both Gaussian and non-Gaussian random fields to investigate the effect of the intrinsic dimensionality of input data. We demonstrate both the advantages and limitations of the unsupervised learning methods and we conclude that a suitable m-PCE model provides a cost-effective approach compared to alternative algorithms proposed in the literature, including recently proposed expensive deep neural network-based surrogates and can be readily applied for high-dimensional UQ in stochastic PDEs.

42 ENGINEERING↗

Phasor-Measurement-Unit-Based Data Analytics Using Digital Twin and PhasorAnalytics Software

A major objective of this project was to apply GE’s commercial machine learning and data analytics toolsets to large-scale, real-world, anonymized Phasor Measurement Unit (PMU) datasets in order to extract signatures, correlated and/or causal factors, and precursor patterns associated with significant power system phenomena. The project had a particular emphasis on extraction of insights relevant to asset health monitoring, real-time load modeling and cybersecurity monitoring. Additionally, the team was directed to undertake a comprehensive data quality analysis for the provided datasets and encouraged to estimate the ‘machine-learning readiness’ of the datasets by documenting any major obstacles to the application of commercial machine learning algorithms. To accomplish the aforementioned objectives, the project team’s work centered around the identification of key event signatures and application of the identified event signatures for event detection and event classification. The industry-validated, semi-supervised machine learning strategy employed for event signature identification involved several major tasks, including data-preprocessing, generation of an overabundance of features, normal data identification, normality modeling, and event signature identification through a methodical, quantitative ranking of features in order of relevance to each studied event type. Throughout the project, data quality issues and mitigation techniques were investigated. In this report, insights are provided regarding the readiness of the provided synchrophasor datasets for application of machine learning and data analytics. The methodologies employed for this technical strategy are summarized in this report. With regards to data preprocessing and feature generation, the provided Training and Test Datasets were ingested into GE’s big data environment. Subsequently, the team applied bad data cleansing and data imputation scripts, event detection scripts, and application programming interfaces (APIs) to the datasets for convenient data access. The project team completed development and validation of dozens of physics-based, statistics-based and transformation-based feature functions used for the extraction of over 60 synchrophasor features. Using a new parallel feature generation technology developed on this project, over 60 features have been rapidly generated for the full two years’ worth of Training and Test Dataset data associated with both the Eastern and Western interconnects. Even accommodating for temporal down-sampling inherent to the feature extraction procedure, this parallel feature generation activity resulted in a massive feature set with a storage requirement approximately equal to that of the raw training dataset itself. With regards to normal data identification and normality modeling, a normality model was built using the feature data extracted from the Training Dataset and iteratively refined subsequent to incremental adjustments and expansions of the Training Dataset feature data. With respect to event characterization and signature identification, an event signature identification pipeline was developed and used in conjunction with the normality model to identify over 15 event signatures for key event categories within the Training Dataset. The identified event signatures were used to characterize hundreds of key events in terms of relative severity, duration, and location of the event. An investigation was undertaken to identify correlated and causal factors involved in transformer events. A separate investigation into temporal trends in ring-down analysis results was undertaken to determine possible associations between system dynamics and various other factors such as loading, season or year. To validate the identified event signatures, additional work was undertaken to develop signature-based anomaly detection and classification tools suitable for convenient application to the synchrophasor datasets. The anomaly detection and classification tools, suitable for online application, were then applied to the entirety of the Eastern Interconnect Training and Test Datasets. Performance of the event detection and classification tools was evaluated upon receipt of the Test Dataset event logs (i.e., the labels for events contained in the Test Dataset), and promising results were obtained despite several challenges (documented herein) associated with application of supervised or semi-supervised machine learning methods to large-scale, anonymized datasets. Finally, the detection and classification tools were used to detect, classify, and characterize thousands of new events not included in the original event logs provided by the DOE within both the Training and Test Datasets.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Machine learning the metastable phase diagram of covalently bonded carbon

Abstract Conventional phase diagram generation involves experimentation to provide an initial estimate of the set of thermodynamically accessible phases and their boundaries, followed by use of phenomenological models to interpolate between the available experimental data points and extrapolate to experimentally inaccessible regions. Such an approach, combined with high throughput first-principles calculations and data-mining techniques, has led to exhaustive thermodynamic databases (e.g. compatible with the CALPHAD method), albeit focused on the reduced set of phases observed at distinct thermodynamic equilibria. In contrast, materials during their synthesis, operation, or processing, may not reach their thermodynamic equilibrium state but, instead, remain trapped in a local (metastable) free energy minimum, which may exhibit desirable properties. Here, we introduce an automated workflow that integrates first-principles physics and atomistic simulations with machine learning (ML), and high-performance computing to allow rapid exploration of the metastable phases to construct “metastable” phase diagrams for materials far-from-equilibrium. Using carbon as a prototypical system, we demonstrate automated metastable phase diagram construction to map hundreds of metastable states ranging from near equilibrium to far-from-equilibrium (400 meV/atom). We incorporate the free energy calculations into a neural-network-based learning of the equations of state that allows for efficient construction of metastable phase diagrams. We use the metastable phase diagram and identify domains of relative stability and synthesizability of metastable materials. High temperature high pressure experiments using a diamond anvil cell on graphite sample coupled with high-resolution transmission electron microscopy (HRTEM) confirm our metastable phase predictions. In particular, we identify the previously ambiguous structure of n -diamond as a cubic-analog of diaphite-like lonsdaelite phase.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

On the Formalization of Development and Assessment Process for Digital Twins in the Nearly Autonomous Management and Control System

In recent years, the autonomous control system has been encouraged in advanced reactors for restoring economic viability, simplifying the operation and maintenance, and enabling remote-site power generations [1]. Since the reactor is expected to be operated for a long period of time with a limited number of individuals onsite, it is recommended that the autonomous control system should have access to very realistic models of the state of processes in the whole lifecycle, together with these process behaviors in interaction with their environment in the real world. As a result, digital twin (DT) technology is suggested in autonomous control systems. DT is defined as a digital representation of a physical object or system, which contains a record for the histories of loads, operation and maintenance status, predictions for the near-term transient of important state variables, and decision-making process [2]. Since machine learning (ML) can recognize patterns within a complex system in real-time applications, it has been used to build DTs in the autonomous control systems for advanced reactors. Meanwhile, due to the rareness of operation data in accident scenarios, the development and assessment of DTs is expected to be mainly driven by simulations. Although the capability and feasibility of ML-based DTs are recognized in improving the safety and efficiency of reactor control, a major concern from the regulatory commission and the nuclear industry is whether the information from a DT is developed and assessed in accordance with expectation and requirements by the target decision. Such concerns not only affect the acceptance criteria for DTs, but also values that can be extracted from DTs and autonomous control system during operations. Inspired by the success of formal methods in improving the reliability and robustness of computer programming and software development, it is suggested that the development and assessment process (DAP) for both separate DTs and integral control system should be formalized in a transparent, consistent, and improvable manner. In this study, a digital-twin development and assessment process (DT-DAP) is proposed by adapting the evaluation model development and assessment process (EMDAP) [3] to requirements by the autonomous control system, ML algorithms, and DT technology. To demonstrate the framework, a baseline nearly autonomous management and control (NAMAC) system with ML-based DTs for diagnosis and prognosis is developed and assessed based on the framework. It is found that with selected testing methods and techniques, the DT-DAP can help identify errors in DTs and NAMAC which would otherwise be left unverified. Meanwhile, it is found that the DT-DAP can improve the DTs and NAMAC by continuously learning and iterating through different elements.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Reducing Urban Traffic Congestion Using Deep Learning and Model Predictive Control

This article proposes a deep learning (DL)-based control algorithm—DL velocity-based model predictive control (VMPC)—for reducing traffic congestion with slowly time-varying traffic signal controls. This control algorithm consists of system identification using DL and traffic signal control using VMPC. For the training process of DL, we established a modeling error entropy loss as the criteria inspired by the theory of stochastic distribution control (SDC) originated by the fourth author. Herein, simulation results show that the proposed algorithm can reduce traffic congestion with a slowly varying traffic signal control input. Results of an ablation study demonstrate that this algorithm compares favorably to other model-based controllers in terms of prediction error, signal varying speed, and control effectiveness.

97 MATHEMATICS AND COMPUTING↗

Predicting disruptive instabilities in controlled fusion plasmas through deep learning

Nuclear fusion power delivered by magnetic-confinement tokamak reactors holds the promise of sustainable and clean energy1. The avoidance of large-scale plasma instabilities called disruptions within these reactors is one of the most pressing challenges, because disruptions can halt power production and damage key components. Disruptions are particularly harmful for large burning-plasma systems such as the multibillion-dollar International Thermonuclear Experimental Reactor (ITER) project currently under construction, which aims to be the first reactor that produces more power from fusion than is injected to heat the plasma. Here we present a method based on deep learning for forecasting disruptions. Our method extends considerably the capabilities of previous strategies such as first-principles-based and classical machine-learning approaches. In particular, it delivers reliable predictions for machines other than the one on which it was trained—a crucial requirement for future large reactors that cannot afford training disruptions. Our approach takes advantage of high-dimensional training data to boost predictive performance while also engaging supercomputing resources at the largest scale to improve accuracy and speed. Trained on experimental data from the largest tokamaks in the United States (DIII-D) and the world (Joint European Torus, JET), our method can also be applied to specific tasks such as prediction with long warning times: this opens up the possibility of moving from passive disruption prediction to active reactor control and optimization. These initial results illustrate the potential for deep learning to accelerate progress in fusion-energy science and, more generally, in the understanding and prediction of complex physical systems.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

WE-F-201-03: Evaluate Clinical Cases Using Commercially Available Systems and Compare to TG-43 Dosimetry

With the recent introduction of heterogeneity correction algorithms for brachytherapy, the AAPM community is still unclear on how to commission and implement these into clinical practice. The recently-published AAPM TG-186 report discusses important issues for clinical implementation of these algorithms. A charge of the AAPM-ESTRO-ABG Working Group on MBDCA in Brachytherapy (WGMBDCA) is the development of a set of well-defined test case plans, available as references in the software commissioning process to be performed by clinical end-users. In this practical medical physics course, specific examples on how to perform the commissioning process are presented, as well as descriptions of the clinical impact from recent literature reporting comparisons of TG-43 and heterogeneity-based dosimetry. Learning Objectives: Identify key clinical applications needing advanced dose calculation in brachytherapy. Review TG-186 and WGMBDCA guidelines, commission process, and dosimetry benchmarks. Evaluate clinical cases using commercially available systems and compare to TG-43 dosimetry.

60 APPLIED LIFE SCIENCES↗

Reinforcement learning based hybrid bond-order coarse-grained interatomic potentials for exploring mesoscale aggregation in liquid–liquid mixtures

Exploring mesoscopic physical phenomena has always been a challenge for brute-force all-atom molecular dynamics simulations. Although recent advances in computing hardware have improved the accessible length scales, reaching mesoscopic timescales is still a significant bottleneck. Coarse-graining of all-atom models allows robust investigation of mesoscale physics with a reduced spatial and temporal resolution but preserves desired structural features of molecules, unlike continuum-based methods. Here, we present a hybrid bond-order coarse-grained forcefield (HyCG) for modeling mesoscale aggregation phenomena in liquid–liquid mixtures. The intuitive hybrid functional form of the potential offers interpretability to our model, unlike many machine learning based interatomic potentials. We parameterize the potential with the continuous action Monte Carlo Tree Search (cMCTS) algorithm, a reinforcement learning (RL) based global optimizing scheme, using training data from all-atom simulations. The resulting RL-HyCG correctly describes mesoscale critical fluctuations in binary liquid–liquid extraction systems. cMCTS, the RL algorithm, accurately captures the mean behavior of various geometrical properties of the molecule of interest, which were excluded from the training set. The developed potential model along with the RL-based training workflow could be applied to explore a variety of other mesoscale physical phenomena that are typically inaccessible to all-atom molecular dynamics simulations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

EPICS for small-scale laboratories with Python soft IOCs

While the Experimental Physics and Industrial Control System (EPICS) is widely used at large laboratories for slow controls and instrumentation, the deployment of a full EPICS installation can be difficult, with a steep learning curve to new users. Taking advantage of the pythonSoftIOC module, we developed an EPICS slow controls implementation for Jefferson Lab's Hall B cryotarget written entirely in Python and based on software IOCs that communicate with instruments over Ethernet. Here, this system ran successfully, interfacing with Jefferson Lab's full EPICS network, and we offer it as an example of the capabilities of pythonSoftIOC to build lightweight, yet robust and flexible instrumentation platforms that would be easily adapted for use at a small-scale laboratory. University groups can use these examples to build complete slow controls systems, from device communication to data archiving and display, using open-source, mature EPICS tools and student-friendly Python as an alternative to expensive and proprietary systems such as LabVIEW.

Computing↗

AI-powered municipal solid waste management: a comprehensive review from generation to utilization

The accumulation of municipal solid waste (MSW) continues to rise due to burgeoning population, rapid global urbanization and economic growth, intensifying ecological concerns associated with landfills and greenhouse gas (GHG) emissions. Over the past 2 decades, global waste generation has surged by 50%, with one-third remaining uncollected and about 70% sent to landfills. This review examines the critical role of integrating emerging technologies, such as advanced sensors and artificial intelligence (AI), into end-to-end MSW management to alleviate landfill burdens. The suitability of various AI tools for different stages of MSW management is assessed, alongside the deployment of advanced sensors including hyperspectral cameras, computer vision systems, and internet of things (IoT) devices for material identification. Applications of genetic algorithms and reinforcement learning for optimizing collection routes, reducing costs, and lowering emissions are highlighted. Life cycle assessment (LCA) across all stages of MSW management is also reviewed, along with future trends in leveraging generative AI, natural language processing (NLP), and agent-based AI systems to analyze waste generation patterns and public sentiment. Efficient collection and handling can be enhanced through route optimization with geographic information systems and real-time bin-level monitoring. Furthermore, sensor-embedded, real-time object detection systems paired with robotics enable material characterization and automated sorting, thereby lowering costs and diverting waste from landfills into value-added products for diverse industrial sectors including packaging, chemicals, textiles, metals and glass, transportation, and electronics industries. Without intervention, global waste is projected to reach 4.54 billion tons by 2050, contributing direct economic costs of $\$$400 billion and roughly 2.38 billion tons of CO 2 -equivalent emissions annually. This review demonstrates how AI-driven, end-to-end solutions for MSW management can mitigate economic and environmental challenges, while directly supporting the United Nations Sustainable Development (UNDP) goals related to innovation and infrastructure (SDG 9), sustainable cities (SDG 11), responsible consumption and production (SDG 12), and climate action (SDG 13).

09 BIOMASS FUELS↗

A Computational Systems Biology Study for Understanding Salt Tolerance Mechanism in Rice

Salinity is one of the most common abiotic stresses in agriculture production. Salt tolerance of rice (Oryza sativa) is an important trait controlled by various genes. The mechanism of rice salt tolerance, currently with limited understanding, is of great interest to molecular breeding in improving grain yield. In this study, a gene regulatory network of rice salt tolerance is constructed using a systems biology approach with a number of novel computational methods. We developed an improved volcano plot method in conjunction with a new machine-learning method for gene selection based on gene expression data and applied the method to choose genes related to salt tolerance in rice. The results were then assessed by quantitative trait loci (QTL), co-expression and regulatory binding motif analysis. The selected genes were constructed into a number of network modules based on predicted protein interactions including modules of phosphorylation activity, ubiquity activity, and several proteinase activities such as peroxidase, aspartic proteinase, glucosyltransferase, and flavonol synthase. All of these discovered modules are related to the salt tolerance mechanism of signal transduction, ion pump, abscisic acid mediation, reactive oxygen species scavenging and ion sequestration. We also predicted the three-dimensional structures of some crucial proteins related to the salt tolerance QTL for understanding the roles of these proteins in the network. Our computational study sheds some new light on the mechanism of salt tolerance and provides a systems biology pipeline for studying plant traits in general.

59 BASIC BIOLOGICAL SCIENCES↗

Fast Machine Learning for Quantum Control of Microwave Qudits on Edge Hardware

Quantum optimal control is a promising approach to improve the accuracy of quantum gates, but it relies on complex algorithms to determine the best control settings. CPU or GPU-based approaches often have delays that are too long to be applied in practice. It is paramount to have systems with extremely low delays to quickly and with high fidelity adjust quantum hardware settings, where fidelity is defined as overlap with a target quantum state. Here, we utilize machine learning (ML) models to determine control-pulse parameters for preparing Selective Number-dependent Arbitrary Phase (SNAP) gates in microwave cavity qudits, which are multi-level quantum systems that serve as elementary computation units for quantum computing. The methodology involves data generation using classical optimization techniques, ML model development, design space exploration, and quantization for hardware implementation. Our results demonstrate the efficacy of the proposed approach, with optimized models achieving low gate trace infidelity near $10^{-3}$ and efficient utilization of programmable logic resources.

Sanders, Flor [Columbia U.]↗

Adaptive Fault Detection Based on Neural Networks and Multiple Sampling Points for Distribution Networks and Microgrids

Smart networks such as microgrid (MG) and active distribution networks (ADN) have been recently playing an important role in power system operation. The design and implementation of appropriate protection systems for such networks must be addressed, which imposes new technical challenges. This paper presents the implementation and validation aspects of an adaptive fault detection strategy based on neural networks (NNs) and multiple sampling points for ADN and MG. The solution is implemented on an edge device. Artificial NNs are used to derive a data-driven model that uses only local measurements to detect fault states of the network without the need for communication infrastructure. Multiple sampling points are used to derive a data-driven model, which allows the generalization considering the implementation in physical systems. The adaptive fault detector model is implemented on a Jetson Nano system, which is a single-board computer (SBC) with a small Graphic Processing Unit (GPU) intended to run machine learning loads at the edge. The proposed method is tested in a physical, real-life, low-voltage network located at Universidad del Norte, Colombia. This testing network is based on the IEEE-13 Node Test Feeder scaled down to 220 V. The validation in a simulation environment shows the accuracy and dependability above 99.6%, while the real-time tests show the accuracy and dependability of 95.5% and 100%, respectively. Without hard-to-derive parameters, the easy-to-implement embedded model highlights the potential for real-life applications.

42 ENGINEERING↗

Enforcing constraints for time series prediction in supervised, unsupervised and reinforcement learning

We assume that we are given a time series of data from a dynamical system and our task is to learn the flow map of the dynamical system. We present a collection of results on how to enforce constraints coming from the dynamical system in order to accelerate the training of deep neural networks to represent the flow map of the system as well as increase their predictive ability. In particular, we provide ways to enforce constraints during training for all three major modes of learning, namely supervised, unsupervised and reinforcement learning. In general, the dynamic constraints need to include terms which are analogous to memory terms in model reduction formalisms. Such memory terms act as a restoring force which corrects the errors committed by the learned flow map during prediction. For supervised learning, the constraints are added to the objective function. For the case of unsupervised learning, in particular generative adversarial networks, the constraints are introduced by augmenting the input of the discriminator. Finally, for the case of reinforcement learning and in particular actor-critic methods, the constraints are added to the reward function. In addition, for the reinforcement learning case, we present a novel approach based on homotopy of the action-value function in order to stabilize and accelerate training. We use numerical results for the Lorenz system to illustrate the various constructions.

Stinis, Panagiotis↗