Search NASA⌕ Search

SEARCH · Search NASA

Results for “Complex Network Data”

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 577 records · Page 32

Networked Microgrid Ownership, Data, and Control Implications: Challenges and Open Questions

Microgrid deployments increasingly favor the potential to form networks for greater benefits to resilience, reliability, and energy sovereignty. Both independent and networked micro-grids predominantly have a single-entity-ownership and control, where the associations from ownership to data requirements to control functions to microgrid objectives is linear. The emerging model, however, is cyclical, with bidirectional causal impacts between each of the 4 pillars: there are more complex mixed ownership models across the physical, electrical, data, communications, protection, and control boundaries that impact the data requirements for meeting control functions that help realize the use-cases or objectives. This paper is the first to delineate the pillars for effective ownership and controllability of both independent as well as networked microgrids through the cyclical model, and present barriers to the adoption of such a model.

Sundararajan, Aditya↗

PROPERTIES OF INTERSTELLAR TURBULENCE FROM GRADIENTS OF LINEAR POLARIZATION MAPS

Faraday rotation of linearly polarized radio signals provides a very sensitive probe of fluctuations in the interstellar magnetic field and ionized gas density resulting from magnetohydrodynamic (MHD) turbulence. We used a set of statistical tools to analyze images of the spatial gradient of linearly polarized radio emission (|{nabla}P|) for both observational data from a test image of the Southern Galactic Plane Survey (SGPS) and isothermal three-dimensional simulations of MHD turbulence. Visually, in both observations and simulations, a complex network of filamentary structures is seen. Our analysis shows that the filaments in |{nabla}P| can be produced both by interacting shocks and random fluctuations characterizing the non-differentiable field of MHD turbulence. The latter dominates for subsonic turbulence, while the former is only present in supersonic turbulence. We show that supersonic and subsonic turbulence exhibit different distributions as well as different morphologies in the maps of |{nabla}P|. Particularly, filaments produced by shocks show a characteristic 'double jump' profile at the sites of shock fronts resulting from delta function-like increases in the density and/or magnetic field, while those produced by subsonic turbulence show a single jump profile. In order to quantitatively characterize these differences, we use the topology tool known as the genus curve as well as the probability distribution function moments of the image distribution. We find that higher values for the moments correspond to cases of |{nabla}P| with larger sonic Mach numbers. The genus analysis of the supersonic simulations of |{nabla}P| reveals a 'swiss cheese' topology, while the subsonic cases have characteristics of a 'clump' topology. Based on the analysis of the genus and the higher order moments, the SGPS test region data have a distribution and morphology that match subsonic- to transonic-type turbulence, which confirms what is now expected for the warm ionized medium.

79 ASTRONOMY AND ASTROPHYSICS↗

A neural network for determination of latent dimensionality in Nonnegative Matrix Factorization

Non-negative Matrix Factorization (NMF) has proven to be a powerful unsupervised learning method for uncovering hidden features in complex and noisy datasets with applications in data mining, text recognition, dimension reduction, face recognition, anomaly detection, blind source separation, and many other fields. An important input for NMF is the latent dimensionality of the data, that is, the number of hidden features, K, present in the explored dataset. Unfortunately, and this quantity is rarely known a priori. The existing methods for determining latent dimensionality, such as Automatic Relevance Determination (ARD), are mostly heuristic and utilize different characteristics to estimate the number of hidden features. However, all of them require human presence to make a final determination of K. Here we utilize a supervised machine learning approach in combination with a recent method for model determination, called NMFk, to determine the number of hidden features automatically. NMFk performs a set of NMF simulations on an ensemble of matrices, obtained by bootstrapping the initial dataset, and estimates which K produces stable groups of latent features that reconstruct the initial dataset well. We then train a Multi-Layter Perceptron (MLP) classifier network to determine the correct number of latent features utilizing the statistics and characteristics of the NMF solution, obtained from NMFk. In order to train the MLP classifier, a training set of 58,660 matrices with predetermined latent features were factorized with NMFk. The MLP classifier in conjunction with NMFk maintains a greater than 95% success rate when applied to a held out test set. Additionally, when applied to two well-known benchmark datasets, the swimmer and MIT face data, NMFk/MLP correctly recovers the established number of hidden features. Finally, we compare the accuracy of our method to the ARD, AIC and Stability-based methods.

97 MATHEMATICS AND COMPUTING↗

Data-driven parameterization of the generalized Langevin equation

Significance The generalized Langevin equation (GLE) provides a precise description of coarse-grained variable dynamics in reduced dimension models. However, computation of the memory kernel poses a major challenge to the practical use of the GLE. This paper presents a data-driven approach to compute the memory kernel, relying on a hierarchy of parameterized rational approximations in terms of the Laplace transform, which can be expanded to arbitrarily high order as necessary. This parameterization makes it convenient to represent the GLE via an extended stochastic model where the memory term is eliminated by properly introducing auxiliary variables. The present method is well-suited for constructing reduced models for nonequilibrium properties of complex systems such as biomolecules, chemical reaction networks, and climate simulations.

97 MATHEMATICS AND COMPUTING↗

Gateway Autonomy for Enabling Deep Space Exploration

The Gateway spacecraft is an important stepping-stone to exploration of the solar system, integrating commercial and international partners into a tightly coupled system, enabling cislunar activities, and implementing key technologies for missions to Mars. Autonomy is a capability area necessary to handle long communication outages where intervention from Earth is impossible, to prepare to operate with long communication delays that will be common in interplanetary travel, and to make spaceflight more affordable and accessible by reducing sustaining operations costs. The Gateway Concept of Operations states that one of Gateway’s goals is to “focus on infrastructure and systems that will allow autonomous operations aboard the Gateway with robotics, automated systems, advanced communications, and distributed computing.” Gateway’s Vehicle Systems Manager (VSM) and associated Autonomous Spacecraft Management Architecture (ASMA) are key products towards delivering autonomous capability. The primary functions of the control architecture are Mission Management and Timeline Execution, Resource Management, Fault Management, and Vehicle Control and Operation (VCO). In each of these areas, there is an initial level of capability to be delivered at launch, with plans to continue development and grow to greater capability. The initial deployment of VSM will focus on maintaining vehicle safety by focusing on full fault management capabilities and deploying only enough resource and timeline planning functionality to support that. The final deployment of VSM will add significant planning and control optimization functionality to support nominal operations for up to 21 days without ground support, even accommodating fault and failure conditions. While the VSM is the vehicle-level representation of autonomous reasoning, distributed automation is essential to provide the right scope and abstraction of information to process. Module and system support of automation and simplicity of interfaces are two important design paradigms that Gateway is focusing on to garner a systems approach to autonomy. Distribution of reasoning can increase complexity, so Gateway is also taking a strict hierarchical approach to information flow and decision making. VSM is not the only capability necessary to achieve an autonomous spacecraft. Robotics support for maintenance of the spacecraft will be essential to provide continued vehicle functionality even when crew is not present. Technical and programmatic challenges exist when implementing autonomous robotics operations. These challenges include sufficient network flexibility to support data transfer to the rest of the vehicle to coordinate module-to-module robotic walk-offs and finding the proper interfaces to allow sufficient dexterity. Communication system upgrades planned for Gateway include Delay Tolerant Networking to best utilize the complex network of relays that will be part of mature cislunar operations. Distributed computing and management will provide failure tolerance, robustness, and growth of capabilities while still allowing significant reuse of heritage software on heritage systems as well as reuse of common applications across a spacecraft to minimize new development, but this requires adherence to key standards and interfaces. The Gateway program has demonstrated significant progress towards these capabilities and has identified challenges other spacecraft developers should be aware of from the start.

Molly Anderson↗

Gateway Autonomy for Enabling Deep Space Exploration

The Gateway spacecraft is an important stepping-stone to exploration of the solar system, integrating commercial and international partners into a tightly coupled system, enabling cislunar activities, and implementing key technologies for missions to Mars. Autonomy is a capability area necessary to handle long communication outages where intervention from Earth is impossible, to prepare to operate with long communication delays that will be common in interplanetary travel, and to make spaceflight more affordable and accessible by reducing sustaining operations costs. The Gateway Concept of Operations states that one of Gateway’s goals is to “focus on infrastructure and systems that will allow autonomous operations aboard the Gateway with robotics, automated systems, advanced communications, and distributed computing.” Gateway’s Vehicle Systems Manager (VSM) and associated Autonomous Spacecraft Management Architecture (ASMA) are key products towards delivering autonomous capability. The primary functions of the control architecture are Mission Management and Timeline Execution, Resource Management, Fault Management, and Vehicle Control and Operation (VCO). In each of these areas, there is an initial level of capability to be delivered at launch, with plans to continue development and grow to greater capability. The initial deployment of VSM will focus on maintaining vehicle safety by focusing on full fault management capabilities and deploying only enough resource and timeline planning functionality to support that. The final deployment of VSM will add significant planning and control optimization functionality to support nominal operations for up to 21 days without ground support, even accommodating fault and failure conditions. While the VSM is the vehicle-level representation of autonomous reasoning, distributed automation is essential to provide the right scope and abstraction of information to process. Module and system support of automation and simplicity of interfaces are two important design paradigms that Gateway is focusing on to garner a systems approach to autonomy. Distribution of reasoning can increase complexity, so Gateway is also taking a strict hierarchical approach to information flow and decision making. VSM is not the only capability necessary to achieve an autonomous spacecraft. Robotics support for maintenance of the spacecraft will be essential to provide continued vehicle functionality even when crew is not present. Technical and programmatic challenges exist when implementing autonomous robotics operations. These challenges include sufficient network flexibility to support data transfer to the rest of the vehicle to coordinate module-to-module robotic walk-offs and finding the proper interfaces to allow sufficient dexterity. Communication system upgrades planned for Gateway include Delay Tolerant Networking to best utilize the complex network of relays that will be part of mature cislunar operations. Distributed computing and management will provide failure tolerance, robustness, and growth of capabilities while still allowing significant reuse of heritage software on heritage systems as well as reuse of common applications across a spacecraft to minimize new development, but this requires adherence to key standards and interfaces. The Gateway program has demonstrated significant progress towards these capabilities and has identified challenges other spacecraft developers should be aware of from the start.

Molly Anderson↗

Lightning Radio Source Retrieval Using Advanced Lightning Direction Finder (ALDF) Networks

A linear algebraic solution is provided for the problem of retrieving the location and time of occurrence of lightning ground strikes from an Advanced Lightning Direction Finder (ALDF) network. The ALDF network measures field strength, magnetic bearing and arrival time of lightning radio emissions. Solutions for the plane (i.e., no Earth curvature) are provided that implement all of tile measurements mentioned above. Tests of the retrieval method are provided using computer-simulated data sets. We also introduce a quadratic planar solution that is useful when only three arrival time measurements are available. The algebra of the quadratic root results are examined in detail to clarify what portions of the analysis region lead to fundamental ambiguities in source location. Complex root results are shown to be associated with the presence of measurement errors when the lightning source lies near an outer sensor baseline of the ALDF network. In the absence of measurement errors, quadratic root degeneracy (no source location ambiguity) is shown to exist exactly on the outer sensor baselines for arbitrary non-collinear network geometries. The accuracy of the quadratic planar method is tested with computer generated data sets. The results are generally better than those obtained from the three station linear planar method when bearing errors are about 2 deg. We also note some of the advantages and disadvantages of these methods over the nonlinear method of chi(sup 2) minimization employed by the National Lightning Detection Network (NLDN) and discussed in Cummins et al.(1993, 1995, 1998).

Koshak, William J.↗

An Adaptive Approach to Energy Security and Assurance

The North American power network has been called the world s largest, most complex machine. This network represents an enormous continent-wide investment, including over 15,000 generators in 10,000 power plants, and miles of transmission lines, distribution networks, and control systems. It has long been recognized that existing power-delivery systems are vulnerable to natural disasters and adversarial attack. The next generation of control systems must simultaneously protect against these disruptions while integrating new generation sources and maintaining active, automated control of the grid. The grid s interconnectedness needed to provide smart grid control also increases its vulnerability to interconnect-wide disruptions that are initiated locally. While strong centralized control is essential to reliable operations, this control requires multiple, high-data-rate, two-way communication links, a powerful central computing facility, and an elaborate operation-control center, all of which present vulnerabilities. It is clearly impossible to completely protect such a widespread and complex system. The key to maintaining the integrity of this network lies in the ability to identify its critical elements and to then effectively protect these elements. This identification and targeted protection will enable parts of the network to remain operational and even automatically reconfigure in the event of local failures or threats of failure, providing dependable continuity of service. The bold concept proposed here is the communication, predictive models and fast controls that eliminate the need to protect all components from emergent contingencies, but to reconfigure actively to maintain critical functions. The transformation from laboratory applications to larger than micro grid areas will require the next generation of energy technologies.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Microwave dielectric behavior of vegetation material

The microwave dielectric behavior of vegetation was examined through the development of theoretical models involving dielectric dispersion by both bound and free water and supported by extensive dielectric measurements conducted over a wide range of conditions. The experimental data were acquired using an open-ended coaxial probe that was developed for sensing the dielectric constant of thin layers of materials, such as leaves, from measurements of the complex reflection coefficient using a network analyzer. The probe system was successfully used to record the spectral variation of the dielectric constant over a wide frequency range extending from 0.5 to 20.4 GHz at numerous temperatures between -40 to +40 C. The vegetation samples were measured over a wide range of moisture conditions. To model the dielectric spectrum of the bound water component of the water included in vegetation, dielectric measurements were made for several sucrose-water solutions as analogs for the situation in vegetation. The results were used in conjunction with the experimental data for leaves to determine some of the constant coefficients in the theoretical models. Two models, both of which provide good fit to the data, are proposed.

Elrayes, Mohamed A.↗

Real-Time event reconstruction for Nuclear Physics Experiments using Artificial Intelligence

Charged track reconstruction is a critical task in nuclear physics experiments, enabling the identification and analysis of particles produced in high-energy collisions. Machine learning (ML) has emerged as a powerful tool for this purpose, addressing the challenges posed by complex detector geometries, high event multiplicities, and noisy data. Traditional methods rely on pattern recognition algorithms like the Kalman filter, but ML techniques, such as neural networks, graph neural networks (GNNs), and recurrent neural networks (RNNs), offer improved accuracy and scalability. By learning from simulated and real detector data, ML models can identify and classify tracks, predict trajectories, and handle ambiguities caused by overlapping or missing hits. Moreover, ML-based approaches can process data in near-real-time, enhancing the efficiency of experiments at large-scale facilities like the Large Hadron Collider (LHC) and Jefferson Lab (JLAB). As detector technologies and computational resources evolve, ML-driven charged track reconstruction continues to push the boundaries of precision and discovery in nuclear physics. In these proceedings, we highlight advancements in charged track identification leveraging Artificial Intelligence within the CLAS12 detector, achieving a notable enhancement in experimental statistics compared to traditional methods. Additionally, we showcase real-time event reconstruction capabilities, including the inference of charged particle properties, such as momentum, direction, and species identification, at speeds matching data acquisition rates. These innovations enable the extraction of physics observables directly from the experiment in real-time.

Gavalian, Gagik (ORCID:0000000267385457)↗

Determining electron temperature for small spherical probes from network analyzer measurements of complex impedance

In earlier work, using a network analyzer, it was shown that collisionless resistance (CR) exists in the sheath of a spherical probe when driven by a small rf signal. The CR is inversely proportional to the plasma density gradient at the location where the applied angular frequency equals the plasma frequency {omega}{sub pe}. Recently, efforts have concentrated on a study of the low-to-intermediate frequency response of the probe to the rf signal. At sufficiently low frequencies, the CR is beyond cutoff, i.e., below the plasma frequency at the surface of the probe. Since the electron density at the probe surface decreases as a function of applied (negative) bias, the CR will extend to lower frequencies as the magnitude of negative bias increases. Therefore to eliminate both CR and ion current contributions, the frequencies presently being considered are much greater than the ion plasma frequency, {omega}{sub pi}, but less than the plasma frequency, {omega}{sub pe}(r{sub 0}), where r{sub 0} is the probe radius. It is shown that, in this frequency regime, the complex impedance measurements made with a network analyzer can be used to determine electron temperature. An overview of the theory is presented along with comparisons to data sets made using three stainless steel spherical probes of different sizes in different experimental environments and different plasma parameter regimes. The temperature measurements made by this method are compared to those made by conventional Langmuir probe sweeps; the method shown here requires no curve fitting as is the usual procedure with Langmuir probes when a Maxwell-Boltzmann electron distribution is assumed. The new method requires, however, a solution of the Poisson equation to determine the approximate sheath dimensions and integrals to determine approximate plasma and sheath inductances. The solution relies on the calculation of impedance for a spherical probe immersed in a collisionless plasma and is based on a simple circuit analogy for the plasma. Finally, the temperatures obtained using this method show reasonable agreement with those obtained using a conventional Langmuir sweep analysis of the spheres.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Probabilistic Modeling of Microbial Metabolic Networks for Integrating Partial Quantitative Knowledge Within the Nitrogen Cycle

Understanding the interactions between microbial communities and their environment sufficiently to predict diversity on the basis of physicochemical parameters is a fundamental pursuit of microbial ecology that still eludes us. However, modeling microbial communities is problematic, because (i) communities are complex, (ii) most descriptions are qualitative, and (iii) quantitative understanding of the way communities interact with their surroundings remains incomplete. One approach to overcoming such complications is the integration of partial qualitative and quantitative descriptions into more complex networks. Here we outline the development of a probabilistic framework, based on Event Transition Graph (ETG) theory, to predict microbial community structure across observed chemical data. Using reverse engineering, we derive probabilities from the ETG that accurately represent observations from experiments and predict putative constraints on communities within dynamic environments. These predictions can feedback into the future development of field experiments by emphasizing the most important functional reactions, and associated microbial strains, required to characterize microbial ecosystems.

59 BASIC BIOLOGICAL SCIENCES↗

The role of HiPPI switches in mass storage systems: A five year prospective

New standards are evolving which provide the foundation for multi-gigabit per second data communication structures. The lowest layer protocols are so generalized that they encourage a wide range of application. Specifically, the ANSI High Performance Parallel Interface (HiPPI) is being applied to computer peripheral attachment as well as general data communication networks. The HiPPI Standards suite and technology products which incorporate the standards are introduced. The use of simple HiPPI crosspoint switches to build potentially complex extended 'fabrics' is discussed in detail. Several near term applications of the HiPPI technology are briefly described with additional attention to storage systems. Finally, some related standards are mentioned which may further expand the concepts above.

Gilbert, T. A.↗

Real-Time Projection to Verify Plan Success During Execution

The Mission Data System provides a framework for modeling complex systems in terms of system behaviors and goals that express intent. Complex activity plans can be represented as goal networks that express the coordination of goals on different state variables of the system. Real-time projection extends the ability of this system to verify plan achievability (all goals can be satisfied over the entire plan) into the execution domain so that the system is able to continuously re-verify a plan as it is executed, and as the states of the system change in response to goals and the environment. Previous versions were able to detect and respond to goal violations when they actually occur during execution. This new capability enables the prediction of future goal failures; specifically, goals that were previously found to be achievable but are no longer achievable due to unanticipated faults or environmental conditions. Early detection of such situations enables operators or an autonomous fault response capability to deal with the problem at a point that maximizes the available options. For example, this system has been applied to the problem of managing battery energy on a lunar rover as it is used to explore the Moon. Astronauts drive the rover to waypoints and conduct science observations according to a plan that is scheduled and verified to be achievable with the energy resources available. As the astronauts execute this plan, the system uses this new capability to continuously re-verify the plan as energy is consumed to ensure that the battery will never be depleted below safe levels across the entire plan.

Wagner, David A.↗

On Development of Three-Dimensional Visualization Capabilities in Glenn Research Center Communication Analysis Suite

With NASA’s upcoming mission to return to the Moon sustainably by 2024 and using that success as a means to step onto the barren world of Mars, it remains more important than ever to conduct research and planning as thoroughly and efficiently as possible. In a mission as complex as landing humans onto another celestial body, a network of orbiting satellites and ground stations must accurately and reliably communicate with each other, enabling crucial data communications throughout the mission. Visualizing this important data communication increases the understanding of the data and can accelerate analyses efforts. The purpose of this software development is to create an interactive visualization with data taken MATLAB® scripts in the GRC Communication Analysis Suite that is easy to understand, can show all necessary data, and display the data accurately. The main types of data to visualize are from the State Propagation, Line of Sight and Dynamic Link Margin scripts. These all show positions and orbits of satellites and ground stations, while the Line of Sight data shows when they have the ability to communicate with each other based on their respective antenna positions and fields of view. Additionally, the Dynamic Link Margin mode color-codes the communication link performance onto the Line of Sight access lines. Visualization requires a graphics language that is easily accessible, has the needed features, and able to easily read data produced by the GRC Communication Analysis Suite MATLAB® scripts. ThreeJS, a graphics library for Web Graphics Library, coded in JavaScript was selected for the visualization. The next part of the software development was to move the data from MATLAB® to the JavaScript. The best way to accomplish this was to implement a MATLAB® function converting the output data of the scripts to a JavaScript Object Notation file. A key part of the development was creating the visualization within JavaScript and ThreeJS to visualize any combination of planets, moons, orbits, satellites, ground stations, line of sight links, and handle future features without changing major parts of the code. The current visualization capability runs directly from MATLAB®, and can dynamically create any scene. This software development currently supports the lunar communications analysis underway by NASA, and can be easily expanded upon in the future to aid any analysis requirements to help plan current and future space missions.

Visualization↗

Using Manufacturing Message Specification for Monitor and Control at Venus

A new approach to the monitor and control of spacecraft tracking systems has been developedbased on the Open Systems Interconnection (OSI) process control standard Manufacturing MessageSpecification (MMS). Station subsystems are interconnected using commercial MMS software tosupport interprocessor communication across a Local Area Network (LAN). Significant cost savingsare realized through the incorporation of commercial Software Control and Data Acquisition(SCADA) packages to support the operator interface. A pilot system has been installed and is inoperation at the Deep Space Network (DSN) experimental Venus complex. The DSN operates a new34-meter beam waveguide antenna (DSS-13) at the Goldstone Venus complex in California. The complex composed of various pieces of equipment with some equipment under computer automated control...

Urista, J.↗

A More Accurate Characterization of UH-60A Pitch Link Loads Using Neural Networks

A more accurate, neural-network-based characterization of the full-scale UH-60A maximum, vibratory pitch link loads (MXVPLL) was obtained. The MXVPLL data were taken from the NASA/Army UH-60A Airloads Program flight test database. This database includes data from level flights, and both simple and "complex" maneuvers. In the present context, a complex maneuver was defined as one which involved simultaneous, non-zero aircraft angle-of-bank (associated with turns) and aircraft pitch-rate (associated with a pull-up or a push-over). The present approach combines physical insight followed by the neural networks application. Since existing load factors do not represent the above-defined complex maneuver, a new, combined load factor ('p resent-load-factor') was introduced. A back-propagation type of neural network with five inputs and one output was used to characterize the UH-60A MXVPLL. The neural network inputs were as follows: rotor advance ratio, aircraft gross weight, rotor RPM, air density ratio, and the present-load-factor. The neural network output was the maximum, vibratory pitch link load (MXVPLL). It was shown that a more accurate characterization of the full-scale flight test pitch link loads can be obtained by combining physical insight with a neural-network-based approach.

Kottapalli, Sesi↗

Measuring the Energy Consumption and Efficiency of Deep Neural Networks: An Empirical Analysis and Design Recommendations

Addressing the "Red-AI" trend of rising energy consumption by large-scale neural networks, this study investigates the measured energy consumption of training various fully connected neural network architectures. We introduce the BUTTER-E dataset, an augmentation to the BUTTER Empirical Deep Learning dataset, containing energy consumption and performance data from 41,129 individual experimental runs spanning 30,582 distinct configurations: 13 datasets, 20 sizes (trainable parameters), 8 "shapes", and 14 depths on both CPUs and GPUs using node-level watt-meters. This dataset reveals the complex relationship between dataset size, network structure, and energy use. Our analysis uncovers a surprising, hardware-mediated non-linear relationship between energy efficiency and network design, challenging the assumption that reducing the number of parameters or FLOPs is the best way to achieve greater energy efficiency. We propose a straightforward and effective energy model that accounts for network size, computing, and memory hierarchy. Highlighting the need for cache-considerate algorithm development, we suggest a codesign approach to energy efficient network, algorithm, and hardware design. This work contributes to the fields of sustainable computing and Green AI, offering practical guidance for creating more energy-efficient neural networks and promoting sustainable AI.

97 MATHEMATICS AND COMPUTING↗