SEARCH · Search NASA

Results for “Constrained learning”

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.

SourceResearch connectionFrom yearThrough year

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

At least 55 records · Page 3

All sources NASA NTRS DOE OSTI

DOE OSTI2025

Neural network approaches for parameterized optimal control

Here, we consider numerical approaches for deterministic, finite-dimensional optimal control problems whose dynamics depend on unknown or uncertain parameters. We seek to amortize the solution over a set of relevant parameters in an offline stage to enable rapid decision-making and be able to react to changes in the parameter in the online stage. To tackle the curse of dimensionality arising when the state and/or parameter are high-dimensional, we represent the policy using neural networks. We compare two training paradigms: First, our model-based approach leverages the dynamics and definition of the objective function to learn the value function of the parameterized optimal control problem and obtain the policy using a feedback form. Second, we use actor-critic reinforcement learning to approximate the policy in a data-driven way. Using an example involving a two-dimensional convection-diffusion equation, which features high-dimensional state and parameter spaces, we investigate the accuracy and efficiency of both training paradigms. While both paradigms lead to a reasonable approximation of the policy, the model-based approach is more accurate and considerably reduces the number of PDE solves.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2025

Constrained or unconstrained? Neural-network-based equation discovery from data

Throughout many fields, practitioners often rely on differential equations to model systems. Yet, for many applications, the theoretical derivation of such equations and/or the accurate resolution of their solutions may be intractable. Instead, recently developed methods, including those based on parameter estimation, operator subset selection, and neural networks, allow for the data-driven discovery of both ordinary and partial differential equations (PDEs), on a spectrum of interpretability. The success of these strategies is often contingent upon the correct identification of representative equations from noisy observations of state variables and, as importantly and intertwined with that, the mathematical strategies utilized to enforce those equations. Specifically, the latter has been commonly addressed via unconstrained optimization strategies. Representing the PDE as a neural network, we propose to discover the PDE (or the associated operator) by solving a constrained optimization problem and using an intermediate state representation similar to a physics-informed neural network (PINN). The objective function of this constrained optimization problem promotes matching the data, while the constraints require that the discovered PDE is satisfied at a number of spatial collocation points. We present a penalty method and a widely used trust-region barrier method to solve this constrained optimization problem, and we compare these methods on numerical examples. Our results on several example problems demonstrate that the latter constrained method outperforms the penalty method, particularly for higher noise levels or fewer collocation points. This work motivates further exploration into using sophisticated constrained optimization methods in scientific machine learning, as opposed to their commonly used, penalty-method or unconstrained counterparts. For both of these methods, we solve these discovered neural network PDEs with classical methods, such as finite difference methods, as opposed to PINNs-type methods relying on automatic differentiation. Here, we briefly highlight how simultaneously fitting the data while discovering the PDE improves the robustness to noise and other small, yet crucial, implementation details.

Data-driven discovery↗

NASA NTRS2019

Understanding Heating in Active Region Cores through Machine Learning. I. Numerical Modeling and Predicted Observables

To adequately constrain the frequency of energy deposition in active region cores in the solar corona, systematic comparisons between detailed models and observational data are needed. In this paper, we describe a pipeline for forward modeling active region emission using magnetic field extrapolations and field-aligned hydrodynamic models. We use this pipeline to predict time-dependent emission from active region NOAA 1158 for low-, intermediate-, and high-frequency nanoflares. In each pixel of our predicted multi-wavelength, time-dependent images, we compute two commonly used diagnostics: the emission measure slope and the time lag. We find that signatures of the heating frequency persist in both of these diagnostics. In particular, our results show that the distribution of emission measure slopes narrows and the mean decreases with decreasing heating frequency and that the range of emission measure slopes is consistent with past observational and modeling work. Furthermore, we find that the time lag becomes increasingly spatially coherent with decreasing heating frequency while the distribution of time lags across the whole active region becomes more broad with increasing heating frequency. In a follow-up paper, we train a random forest classifier on these predicted diagnostics and use this model to classify real observations of NOAA 1158 in terms of the underlying heating frequency.

UV radiation↗

NASA NTRS1986

KBGIS-2: A knowledge-based geographic information system

The architecture and working of a recently implemented knowledge-based geographic information system (KBGIS-2) that was designed to satisfy several general criteria for the geographic information system are described. The system has four major functions that include query-answering, learning, and editing. The main query finds constrained locations for spatial objects that are describable in a predicate-calculus based spatial objects language. The main search procedures include a family of constraint-satisfaction procedures that use a spatial object knowledge base to search efficiently for complex spatial objects in large, multilayered spatial data bases. These data bases are represented in quadtree form. The search strategy is designed to reduce the computational cost of search in the average case. The learning capabilities of the system include the addition of new locations of complex spatial objects to the knowledge base as queries are answered, and the ability to learn inductively definitions of new spatial objects from examples. The new definitions are added to the knowledge base by the system. The system is currently performing all its designated tasks successfully, although currently implemented on inadequate hardware. Future reports will detail the performance characteristics of the system, and various new extensions are planned in order to enhance the power of KBGIS-2.

Smith, T.↗

NASA NTRS1986

KBGIS-II: A knowledge-based geographic information system

The architecture and working of a recently implemented Knowledge-Based Geographic Information System (KBGIS-II), designed to satisfy several general criteria for the GIS, is described. The system has four major functions including query-answering, learning and editing. The main query finds constrained locations for spatial objects that are describable in a predicate-calculus based spatial object language. The main search procedures include a family of constraint-satisfaction procedures that use a spatial object knowledge base to search efficiently for complex spatial objects in large, multilayered spatial data bases. These data bases are represented in quadtree form. The search strategy is designed to reduce the computational cost of search in the average case. The learning capabilities of the system include the addition of new locations of complex spatial objects to the knowledge base as queries are answered, and the ability to learn inductively definitions of new spatial objects from examples. The new definitions are added to the knowledge base by the system. The system is performing all its designated tasks successfully. Future reports will relate performance characteristics of the system.

Smith, Terence↗

DOE OSTI2026

Constraining microphysical processes of warm rain formulation using advanced spectral separations, an ensemble retrieval framework and machine learning techniques

Drizzle, a common feature of marine boundary layer clouds formed through collision coalescence, plays a key role in cloud microphysics and evolution. Yet, simultaneously retrieving cloud and drizzle properties from remote-sensing observations remains challenging because drizzle droplets often dominate radar signals, masking cloud contributions. The goal of the proposed research is to provide constraints for the process of autoconversion and accretion using ARM cloud measurements. Specifically, we provide concurrent retrievals of cloud and drizzle that allows users to derive corresponding autoconversion and accretion rates.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2026

An information-matching approach to optimal experimental design and active learning

The efficacy of mathematical models heavily depends on the quality of the training data, yet collecting sufficient data is often expensive and challenging. Many modeling applications require inferring parameters only as a means to predict other quantities of interest (QoI). Because models often contain many unidentifiable (sloppy) parameters, QoIs often depend on a relatively small number of parameter combinations. Therefore, we introduce an information-matching criterion based on the Fisher information matrix to select the most informative training data from a candidate pool. This method ensures that the selected data contain sufficient information to learn only those parameters that are needed to constrain downstream QoIs. It is formulated as a convex optimization problem, making it scalable to large models and datasets. Here, we demonstrate the effectiveness of this approach across various modeling problems in diverse scientific fields, including power systems and underwater acoustics. Finally, we use information-matching as a query function within an active learning (AL) loop for materials science applications. In all these applications, we find that a relatively small set of optimal training data can provide the necessary information for achieving precise predictions. These results are encouraging for diverse future applications, particularly AL in large machine-learning models.

Materials science↗

DOE OSTI2026

Predicting Atomistic Transitions with Transformers

Accurate knowledge of the atomistic transition pathways in materials and material surfaces is crucial for many material science problems. However, conventional simulation techniques used to find these transitions are extremely computationally intensive. Even with large-scale, accelerated material simulations, the computational cost constrains the applicable domain in practice. Machine learning models, with the potential to learn the complex emergent behaviors governing atomistic transitions as a fast surrogate model, have great promise to predict transitions with a vastly reduced computational cost. Here, we demonstrate how transformers can be trained to predict atomistic transitions in nano-clusters. We show how we evaluate physical validity of the predictions and how a multitude of additional, different microstates can be generated by slightly varying the data provided to the model.

36 MATERIALS SCIENCE↗

NASA NTRS2020

Deep Learning Models for Planetary Seismicity Detection

Research in planetary seismology is fundamentally constrained by a lack of data. Seismo-logical science products of future missions can typically only be informed by theoretical signal/noise characteristics of the environment or likely Earth-analogues. Although objectives can be re-assessed after some initial data-collection upon lander arrival, transfer of high-resolution data back to Earth is costly on lander power usage. Over the last several years, development of GPU computing techniques and open-source high-level APIs have led to rapid advances in deep learning within the fields of computer vision, natural language processing, and collaborative filtering. These techniques are actively being adapted in seismology for a variety of tasks, including: earthquake detection, seismic phase discrimination, and ground-motion prediction. Until the recent detection of mars quakes during the Mars InSight mission, the only other measurements of seismicity recorded outside of Earth was on the Moon during the Apollo missions between 1969 to 1977. These unique data sets have been periodically revisited using new seismological methods, including ambient noise interferometry and Hidden Markov Models. Our objective is to develop a deep learning seismic detector and use it to catalog moonquakes from the Apollo 17 Lunar Seismic Profiling Experiment (LSPE) and compare the results with those obtained by other methods. Additionally, we will assess the accuracy tradeoff between using a training set of lunar data and one composed of Earth seismicity. In this document, we present preliminary results using a prototype classifier trained on a small set of earthquakes that was able to obtain detections for LSPE moonquakes with a greater accuracy than a recent study using Hidden Markov Models.

Civilini, F.↗

NASA NTRS1990

Application of adjoint operators to neural learning

A technique for the efficient analytical computation of such parameters of the neural architecture as synaptic weights and neural gain is presented as a single solution of a set of adjoint equations. The learning model discussed concentrates on the adiabatic approximation only. A problem of interest is represented by a system of N coupled equations, and then adjoint operators are introduced. A neural network is formalized as an adaptive dynamical system whose temporal evolution is governed by a set of coupled nonlinear differential equations. An approach based on the minimization of a constrained neuromorphic energylike function is applied, and the complete learning dynamics are obtained as a result of the calculations.

Barhen, J.↗

NASA NTRS2007

Automated Knowledge Discovery From Simulators

A computational method, SimLearn, has been devised to facilitate efficient knowledge discovery from simulators. Simulators are complex computer programs used in science and engineering to model diverse phenomena such as fluid flow, gravitational interactions, coupled mechanical systems, and nuclear, chemical, and biological processes. SimLearn uses active-learning techniques to efficiently address the "landscape characterization problem." In particular, SimLearn tries to determine which regions in "input space" lead to a given output from the simulator, where "input space" refers to an abstraction of all the variables going into the simulator, e.g., initial conditions, parameters, and interaction equations. Landscape characterization can be viewed as an attempt to invert the forward mapping of the simulator and recover the inputs that produce a particular output. Given that a single simulation run can take days or weeks to complete even on a large computing cluster, SimLearn attempts to reduce costs by reducing the number of simulations needed to effect discoveries. Unlike conventional data-mining methods that are applied to static predefined datasets, SimLearn involves an iterative process in which a most informative dataset is constructed dynamically by using the simulator as an oracle. On each iteration, the algorithm models the knowledge it has gained through previous simulation trials and then chooses which simulation trials to run next. Running these trials through the simulator produces new data in the form of input-output pairs. The overall process is embodied in an algorithm that combines support vector machines (SVMs) with active learning. SVMs use learning from examples (the examples are the input-output pairs generated by running the simulator) and a principle called maximum margin to derive predictors that generalize well to new inputs. In SimLearn, the SVM plays the role of modeling the knowledge that has been gained through previous simulation trials. Active learning is used to determine which new input points would be most informative if their output were known. The selected input points are run through the simulator to generate new information that can be used to refine the SVM. The process is then repeated. SimLearn carefully balances exploration (semi-randomly searching around the input space) versus exploitation (using the current state of knowledge to conduct a tightly focused search). During each iteration, SimLearn uses not one, but an ensemble of SVMs. Each SVM in the ensemble is characterized by different hyper-parameters that control various aspects of the learned predictor - for example, whether the predictor is constrained to be very smooth (nearby points in input space lead to similar output predictions) or whether the predictor is allowed to be "bumpy." The various SVMs will have different preferences about which input points they would like to run through the simulator next. SimLearn includes a formal mechanism for balancing the ensemble SVM preferences so that a single choice can be made for the next set of trials.

Burl, Michael↗

DOE OSTI2024

Metric Learning to Accelerate Convergence of Operator Splitting Methods

Recent developments in machine learning have led to promising advances in accelerating the solution of constrained optimization problems. Increasing demand for real-time decision-making capabilities in applications such as artificial intelligence and optimal control has led to a variety of proposed strategies for learning to produce fast solutions to optimization problems. For example, recent works have shown that it is possible to accelerate the convergence of optimization algorithms by learning to select their parameters, such as gradient descent stepsizes. This work proposes a new approach, in which the underlying metric spaces of proximal operator splitting algorithms are learned to maximize convergence rate. While prior works in optimization theory have derived optimal metrics in simple cases, no such result exists for many practical problem forms including general Quadratic Programming (QP). This paper shows how differentiable optimization can enable the end-to-end learning of proximal metrics, enhancing the convergence of proximal algorithms for QP problems beyond what is possible based on known theory. Additionally, the results illustrate a strong connection between the learned proximal metrics and active constraints at the optima, leading to an interpretation in which the predicted proximal metrics can be viewed as a form of active set prediction.

King, Ethan [BATTELLE (PACIFIC NW LAB)]↗

NASA NTRS1992

Machine Learning Techniques in Optimal Design

Many important applications can be formalized as constrained optimization tasks. For example, we are studying the engineering domain of two-dimensional (2-D) structural design. In this task, the goal is to design a structure of minimum weight that bears a set of loads. A solution to a design problem in which there is a single load (L) and two stationary support points (S1 and S2) consists of four members, E1, E2, E3, and E4 that connect the load to the support points is discussed. In principle, optimal solutions to problems of this kind can be found by numerical optimization techniques. However, in practice [Vanderplaats, 1984] these methods are slow and they can produce different local solutions whose quality (ratio to the global optimum) varies with the choice of starting points. Hence, their applicability to real-world problems is severely restricted. To overcome these limitations, we propose to augment numerical optimization by first performing a symbolic compilation stage to produce: (a) objective functions that are faster to evaluate and that depend less on the choice of the starting point and (b) selection rules that associate problem instances to a set of recommended solutions. These goals are accomplished by successive specializations of the problem class and of the associated objective functions. In the end, this process reduces the problem to a collection of independent functions that are fast to evaluate, that can be differentiated symbolically, and that represent smaller regions of the overall search space. However, the specialization process can produce a large number of sub-problems. This is overcome by deriving inductively selection rules which associate problems to small sets of specialized independent sub-problems. Each set of candidate solutions is chosen to minimize a cost function which expresses the tradeoff between the quality of the solution that can be obtained from the sub-problem and the time it takes to produce it. The overall solution to the problem, is then obtained by solving in parallel each of the sub-problems in the set and computing the one with the minimum cost. In addition to speeding up the optimization process, our use of learning methods also relieves the expert from the burden of identifying rules that exactly pinpoint optimal candidate sub-problems. In real engineering tasks it is usually too costly to the engineers to derive such rules. Therefore, this paper also contributes to a further step towards the solution of the knowledge acquisition bottleneck [Feigenbaum, 1977] which has somewhat impaired the construction of rulebased expert systems.

Cerbone, Giuseppe↗

DOE OSTI2026

Decoding the proton’s gluonic density with lattice QCD-informed machine learning

We present a first machine learning-based decoding of the gluonic structure of the proton from lattice QCD using a variational autoencoder inverse mapper (VAIM). Harnessing the power of generative AI, we predict the parton distribution function (PDF) of the gluon given information on the reduced pseudo-Ioffe-time distributions (RpITDs) as calculated from an ensemble with lattice spacing a ≈ 0.09 fm and a pion mass of M π ≈ 310 MeV. The resulting gluon PDF is consistent with phenomenological global fits within uncertainties, particularly in the intermediate-to-high-x region where lattice data are most constraining. A subsequent correlation analysis confirms that the VAIM learns a meaningful latent representation, highlighting the potential of generative AI to bridge lattice QCD and phenomenological extractions within a unified analysis framework.

Gluon parton distribution function↗

DOE OSTI2025

A new biogeochemical modelling framework (FLaMe-v1.0) for lake methane emissions on the regional scale: development and application to the European domain

This study presents a new physical-biogeochemical modelling framework for simulating lake methane (CH 4 ) emissions at regional scales. The new model, FLaMe-v1.0 (Fluxes of Lake Methane), rests on an innovative, computationally efficient lake clustering approach that enables the simulation of CH 4 emissions across a large number of lakes. Building on the Canadian Small Lake Model (CSLM) that simulates the lake physics, we develop a suite of biogeochemical modules to simulate transient dynamics of organic Carbon (C), Oxygen (O 2 ), and CH 4 . We first test the performance of FLaMe-v1.0 by analyzing physical and biogeochemical processes in two theoretical lakes with characteristics that can be considered representative for many lakes (an oligotrophic, deep lake driven by cold climate versus a eutrophic, shallow lake driven by warm climate). Next, we evaluate the model by comparing simulated and observed timeseries of CH 4 emissions in four well-surveyed lakes. We then apply FLaMe-v1.0 at the European scale to evaluate simulated diffusive and ebullitive lake CH 4 fluxes against in-situ measurements in both boreal and central European regions. Finally, we provide a first assessment of the spatio-temporal variability in CH 4 emissions from European lakes with a surface area comprised between 0.1–1000 km 2 (n= 108 407, total area = 1.33 × 105 km 2 ), indicating a total emission of 0.97 ± 0.23 Tg CH 4 yr −1 , with the uncertainty constrained by combining FLaMe-v1.0 and machine learning techniques. Moreover, 30 % and 70 % of these CH 4 emissions are through diffusive and ebullitive pathways, respectively. Annually averaged CH 4 emission rates per unit lake area during 2010–2016 have a South-to-North decreasing gradient, resulting in a mean over the European domain as 7.39 g CH 4 m −2 yr −1 . Our simulations reveal a strong seasonality (with ice-blocking effects accounted for) in European lake CH 4 emissions, with nearly ten times higher emissions during late summer than during winter. This pronounced seasonal variation highlights the importance of accounting for the sub-annual variability in CH 4 emissions to accurately constrain regional CH 4 budgets. In the future, FLaMe-v1.0 could be embedded into Earth System Models to investigate the feedback between climate warming and global lake CH 4 emissions.

Maisonnier, Manon [Free Univ. of Brussels (Belgium↗

DOE OSTI2025

Safe Reinforcement Learning-Based Transient Stability Control for Islanded Microgrids With Topology Reconfiguration

This paper proposes a safe reinforcement learning (RL)-based transient stability emergency control (TSEC) method for islanded microgrids. RL requires extensive interaction with the environment to learn control strategies, hence, a data-driven approach is used as a substitute for time-consuming time-domain simulation calculations. Deep sigma point processes (DSPP), which is a Gaussian process model, is utilized to predict the normal distribution of transient stability of microgrids and to construct a transient stability chance constraint. Reward-constrained policy optimization (RCPO) can simultaneously achieve objective prediction, policy learning, and constraint cost coefficient update across multiple timescales. RCPO interacts with the DSPP-based microgrid environment through a multi-process parallel manner, greatly increasing the training speed. Case studies on a real islanded microgrid demonstrate that the proposed method can efficiently and quickly obtain the optimal emergency control strategy while adhering to all hard constraints.

14 SOLAR ENERGY↗

DOE OSTI2026

Beyond sequence similarity: toward function-based screening of nucleic acid synthesis

Synthetic nucleic acids are a key input to modern biotechnology, yet they represent dual-use materials that require robust screening to mitigate biosecurity risks. The prevailing screening paradigm, which identifies sequences of concern (SoCs) through sequence similarity to controlled pathogens and toxins, may not fully capture risks posed by AI tools that can decouple biomolecular function from reliance on known sequences. Rapidly advancing biodesign capabilities enable the generation of genes and proteins that might evade sequence-based detection. We highlight the critical need for function-based screening approaches that can detect sequences capable of hazardous biological functions, regardless of similarity to known SoCs. We examine the feasibility of function-based screening with an initial focus on proteins, arguing that, while protein sequence space is vast, biologically functional proteins are significantly constrained by biophysical and biochemical requirements that can be learned and modeled. We propose a concrete implementation framework organized along a continuum of complexity, starting with toxins as the most tractable targets before expanding to more complex pathogenic functions. We then discuss open challenges and describe a research and development strategy to address them.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2024

Machine learning-guided discovery of polymer membranes for CO 2 separation with genetic algorithm

Designing polymer membranes with high gas permeability and selectivity is a difficult multi-task constrained problem due to the trade-off between these two properties. In this work, we present a machine learning (ML) driven genetic algorithm to tackle the design problem of polymer membranes for CO 2 separation from N 2 and O 2 . Using literature data of permeability for three gases, we constructed multiple ML models with different fingerprinting featurization schemes to predict gas permeabilities. Then, we employed a genetic algorithm to design new polymers and evaluated their performance using our ML models. We were able to identify new polymer membranes that are promising for both CO 2 /N 2 and CO 2 /O 2 separations. Further, the top discovered polymers are predicted to have high glass transition temperatures. Similarly, the pyridine functionality was found in ≈20% of the predicted polymers. This framework can be used to design polymers for any application involving constrained optimization. Finally, we outlined the challenges and opportunities with using ML guided data-driven inverse design of polymers.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗