Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “generative machine 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.

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

At least 1,657 records · Page 92

A knowledge-informed large language model framework for U.S. nuclear power plant shutdown initiating event classification for probabilistic risk assessment

Identifying and classifying shutdown initiating events (SDIEs) is critical for developing shutdown probabilistic risk assessment for nuclear power plants. Existing computational approaches cannot achieve satisfactory performance due to the challenges of unavailable large, labeled datasets, imbalanced event types, and label noise. To address these challenges, we propose a hybrid pipeline that integrates a knowledge-informed machine learning model to prescreen non-SDIEs and a large language model (LLM) to classify SDIEs into four types. In the prescreening stage, we proposed a set of 44 SDIE text patterns that consist of the most salient keywords and phrases from six SDIE types. Text vectorization based on the SDIE patterns generates feature vectors that are highly separable by using a simple binary classifier. The second stage builds Bidirectional Encoder Representations from Transformers (BERT)-based LLM, which learns generic English language representations from self-supervised pretraining on a large dataset and adapts to SDIE classification by fine-tuning it on an SDIE dataset. The proposed approaches are evaluated on a dataset with 10,928 events using precision, recall ratio, F 1 score, and average accuracy. In conclusion, the results demonstrate that the prescreening stage can exclude more than 97% non-SDIEs, and the LLM achieves an average accuracy of 95.1% for SDIE classification.

99 - GENERAL AND MISCELLANEOUS↗

A general mechanistic framework for cross-scale understanding of hot spots and hot moments in carbon and water fluxes

Semi-arid ecosystems, like those in the American Southwest, exert a massive impact on the interannual variability of carbon and water cycling. Unfortunately, these carbon and water fluxes are notoriously difficult to predict due to their high spatial and temporal variability, which is poorly captured by the current generation of vegetation models. Indeed, this region is exemplified by the ‘hot spots and hot moments’ concept, which states that small areas in space (‘hot spots’) and transient moments in time (‘hot moments’) exert an outsized influence on biogeochemical cycling. However, the factors that regulate these pulses in biogeochemical activity are unknown, as is their variability across space and time. These uncertainties severely limit efforts to better represent hot spots and hot moments in models. Here, we seek to develop a generalized method for detecting and quantifying the importance of hot spots and hot moments from individual plant to regional scales. Underpinning this method is our recently developed statistical approach for identifying hot spots and hot moments. By applying this method to semi-continuous measurements of plant water status, a depth profile of soil water potential, and ecosystem fluxes via eddy covariance, we will track the fate of water through the soil-plant-atmosphere continuum and identify the mechanistic drivers of these transient pulses in biogeochemical activity. Then, we will expand this approach across a broad network of Ameriflux towers, and apply a machine learning approach that will allow us to upscale measurements of hot spots and hot moments across the American Southwest and quantify their impact on carbon and water cycles. These products will allow us to identify hot spots and hot moments across spatio-temporal scales and will serve as crucial data sources for validating a new generation of models that can better capture highly dynamic carbon and water fluxes. The proposed method will be easily transferable across biomes and will serve as a framework for future research on hot spots and hot moments across the plant ecophysiology, biometeorology, and vegetation modeling communities.

54 ENVIRONMENTAL SCIENCES↗

A novel approach for large-scale wind energy potential assessment

Increasing wind energy generation is central to grid decarbonization, yet methods to estimate wind energy potential are not standardized, leading to inconsistencies and even skewed results. This study aims to improve the fidelity of wind energy potential estimates through an approach that integrates geospatial analysis and machine learning (i.e., Gaussian process regression). We demonstrate this approach to assess the spatial distribution of wind energy capacity potential in the Contiguous United States (CONUS). We find that the capacity-based power density ranges from 1.70 MW/km2 (25th percentile) to 3.88 MW/km2 (75th percentile) for existing wind farms in the CONUS. The value is lower in agricultural areas (2.73 ± 0.02 MW/km2, mean ± 95 % confidence interval) and higher in other land cover types (3.30 ± 0.03 MW/km2). Notably, advancements in turbine manufacturing could reduce power density in areas with lower wind speeds by adopting low specific-power turbines, but improve power density in areas with higher wind speeds (>8.35 m/s at 120m above the ground), highlighting opportunities for repowering existing wind farms. Wind energy potential is shaped by wind resource quality and is regionally characterized by land cover and physical conditions, revealing significant capacity potential in the Great Plains and Upper Texas. The results indicate that areas previously identified as hot spots using existing approaches (e.g., the west of the Rocky Mountains) may have a limited capacity potential due to low wind resource quality. Improvements in methodology and capacity potential estimates in this study could serve as a new basis for future energy systems analysis and planning.

Dai, Tao↗

Hydrological Perspectives on Integrated, Coordinated, Open, Networked (ICON) Science

Hydrologic sciences depend on data monitoring, analyses, and simulations of hydrologic processes to ensure safe, sufficient, and equal water distribution. These hydrologic data come from but are not limited to primary (lab, plot, and field experiments) and secondary sources (remote sensing, UAVs, hydrologic models) that typically follow FAIR Principles (Findable, Accessible, Interoperable, and Reusable: (go-fair.org)). Easy availability of FAIR data has become possible because the hydrology-oriented organizations have pushed the community to increase coordination of the protocols for generating data and sharing model platforms. In addition, networking at all levels has emerged with an invigorated effort to activate community science efforts that complement conventional data collection methods. However, it has become difficult to decipher various complex hydrologic processes with increasing data. Machine learning, a branch of artificial intelligence, provide more accurate and faster alternatives to better understand different hydrological processes. The Integrated, Coordinated, Open, Networked (ICON) framework provides a pathway for water users to include and respect diversity, equity, and inclusivity. In addition, ICONs support the integration of peoples with historically marginalized identities into this professional discipline of water sciences. This article comprises three independent commentaries about the state of ICON principles in hydrology and discusses the opportunities and challenges of adopting them.

(ICON) principles to address↗

State predictive information bottleneck

We report the ability to make sense of the massive amounts of high-dimensional data generated from molecular dynamics simulations is heavily dependent on the knowledge of a low-dimensional manifold (parameterized by a reaction coordinate or RC) that typically distinguishes between relevant metastable states, and which captures the relevant slow dynamics of interest. Methods based on machine learning and artificial intelligence have been proposed over the years to deal with learning such low-dimensional manifolds, but they are often criticized for a disconnect from more traditional and physically interpretable approaches. To deal with such concerns, in this work we propose a deep learning based state predictive information bottleneck approach to learn the RC from high-dimensional molecular simulation trajectories. We demonstrate analytically and numerically how the RC learnt in this approach is connected to the committor in chemical physics and can be used to accurately identify transition states. A crucial hyperparameter in this approach is the time delay or how far into the future the algorithm should make predictions about. Through careful comparisons for benchmark systems, we demonstrate that this hyperparameter choice gives useful control over how coarse-grained we want the metastable state classification of the system to be. We thus believe that this work represents a step forward in systematic application of deep learning based ideas to molecular simulations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Machine learned interatomic potential for dispersion strengthened plasma facing components

Tungsten (W) is a material of choice for the divertor material due to its high melting temperature, thermal conductivity, and sputtering threshold. However, W has a very high brittle-to-ductile transition temperature, and at fusion reactor temperatures (≥1000 K), it may undergo recrystallization and grain growth. Dispersion-strengthening W with zirconium carbide (ZrC) can improve ductility and limit grain growth, but much of the effects of the dispersoids on microstructural evolution and thermomechanical properties at high temperatures are still unknown. We present a machine learned Spectral Neighbor Analysis Potential for W–ZrC that can now be used to study these materials. In order to construct a potential suitable for large-scale atomistic simulations at fusion reactor temperatures, it is necessary to train on ab initio data generated for a diverse set of structures, chemical environments, and temperatures. Further accuracy and stability tests of the potential were achieved using objective functions for both material properties and high temperature stability. Additionally, validation of lattice parameters, surface energies, bulk moduli, and thermal expansion is confirmed on the optimized potential. Tensile tests of W/ZrC bicrystals show that although the W(110)–ZrC(111) C-terminated bicrystal has the highest ultimate tensile strength (UTS) at room temperature, observed strength decreases with increasing temperature. At 2500 K, the terminating C layer diffuses into the W, resulting in a weaker W–Zr interface. Meanwhile, the W(110)–ZrC(111) Zr-terminated bicrystal has the highest UTS at 2500 K.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Design of intrinsically disordered protein variants with diverse structural properties

Intrinsically disordered proteins (IDPs) perform a broad range of functions in biology, suggesting that the ability to design IDPs could help expand the repertoire of proteins with novel functions. Computational design of IDPs with specific conformational properties has, however, been difficult because of their substantial dynamics and structural complexity. We describe a general algorithm for designing IDPs with specific structural properties. We demonstrate the power of the algorithm by generating variants of naturally occurring IDPs that differ in compaction, long-range contacts, and propensity to phase separate. We experimentally tested and validated our designs and analyzed the sequence features that determine conformations. We show how our results are captured by a machine learning model, enabling us to speed up the algorithm. Our work expands the toolbox for computational protein design and will facilitate the design of proteins whose functions exploit the many properties afforded by protein disorder.

Science & Technology - Other Topics↗

Optimization of WAAM Process to Produce AUSC Components with Increased Service Life

Additive manufacturing has the potential to revolutionize industrial hardware and unlock efficiency gains through the fabrication of geometries and architectures not possible by conventional processing. Wire Arc Additive manufacturing (WAAM) process is a class of directed energy deposition process enabling higher build rate and allowing custom wire feedstock allowing spatial variation of microstructure. The larger spot size and lower speed creates a larger melt pool, reducing residual stress and often time creating directional/columnar microstructure for Ni-superalloy. The use of flexible platform provides freedom in deposition strategy, which can accommodate complex substrates, including feature addition onto existing structures, non-flat layers, and repair methods. However, the certification of the final component needs to match the strength requirement. Hence, the quality of the component produced is stringently monitored to avoid buildup of residual stress, cracks, porosity and to reduce detrimental segregated phases commonly observed during alloy solidification. Simulation plays a huge role in predicting the melt pool dimension and can be used to optimize the process parameter. Similar development is also required to perform physics-based modeling of microstructural development in WAAM that can predict the microstructural features during solidification and can be used to optimize the process more effectively. To move toward this goal, Raytheon Technologies Research Center together with Siemens worked to create a set of computational tools to control the process parameters, enable on-line measurements and acquisition with feedback to the optimized process parameters, and eventually track material evolution through each step of the additive process. Computational fluid dynamics is used for accurate prediction and calibration of the thermal field during WAAM process. and phase field models for microstructure evolution as a function of processing parameters to establish a connection between additive parameters and the final microstructure. Here we report cellular automata (CA) model development to predict the dendritic microstructure evolution with surface and bulk nuclei for single track and multiple layers. The CA model was developed to account for secondary element addition and predict segregation, local melting, and latent heat release as well as prediction of Euler angles from orientation information and validated against experiments. This framework was utilized to tailor spatially-varying composition in a part by appropriately controlling the microstructure evolution during the additive process. Functionally graded Haynes 282 alloy with high Cr content at the surface was tested for oxidation and mechanical properties. An advanced physics-based reaction-diffusion model predicting the simultaneous creation of chromium oxide and alumina is developed and validated to extend life expectancy of the WAAM manufactured high temperature part. A machine-learning data-driven framework establishing the process-structure relationship from a dataset of real microstructure images and corresponding process history data has been developed and implemented in the NX Siemens design system. The digital twin configuration along with the tool path generation enabled prediction of WAAM component buildup time and the techno-economic analysis provided a favorable option for all 4 cases with 15-40% cost reduction.

33 ADVANCED PROPULSION SYSTEMS↗

Many Molecular Properties from One Kernel in Chemical Space

We introduce property-independent kernels for machine learning modeling of arbitrarily many molecular properties. The kernels encode molecular structures for training sets of varying size, as well as similarity measures sufficiently diffuse in chemical space to sample over all training molecules. Corresponding molecular reference properties provided, they enable the instantaneous generation of ML models which can systematically be improved through the addition of more data. This idea is exemplified for single kernel based modeling of internal energy, enthalpy, free energy, heat capacity, polarizability, electronic spread, zero-point vibrational energy, energies of frontier orbitals, HOMOLUMO gap, and the highest fundamental vibrational wavenumber. Models of these properties are trained and tested using 112 kilo organic molecules of similar size. Resulting models are discussed as well as the kernels’ use for generating and using other property models.

Ramakrishnan, Raghunathan↗

A Systems Biology Approach to Identify Essential Epigenetic Regulators for Specific Biological Processes in Plants

Upon sensing developmental or environmental cues, epigenetic regulators transform the chromatin landscape of a network of genes to modulate their expression and dictate adequate cellular and organismal responses. Knowledge of the specific biological processes and genomic loci controlled by each epigenetic regulator will greatly advance our understanding of epigenetic regulation in plants. To facilitate hypothesis generation and testing in this domain, we present EpiNet, an extensive gene regulatory network (GRN) featuring epigenetic regulators. EpiNet was enabled by (i) curated knowledge of epigenetic regulators involved in DNA methylation, histone modification, chromatin remodeling, and siRNA pathways; and (ii) a machine-learning network inference approach powered by a wealth of public transcriptome datasets. We applied GENIE3, a machine-learning network inference approach, to mine public Arabidopsis transcriptomes and construct tissue-specific GRNs with both epigenetic regulators and transcription factors as predictors. The resultant GRNs, named EpiNet, can now be intersected with individual transcriptomic studies on biological processes of interest to identify the most influential epigenetic regulators, as well as predicted gene targets of the epigenetic regulators. We demonstrate the validity of this approach using case studies of shoot and root apical meristem development.

root apical meristem↗

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.]↗

Where IMERG Goes Next: Version 08 and Beyond

With the Version 07 (V07) Integrated Multi-satellitE Retrievals for GPM (IMERG) algorithm finalized and production initiated, the focus turns to enhancements for Version 08. These include innovations not included in V07 due to time constraints, plus issues revealed by the initial V07 products. One high priority is to evaluate and revise the schemes in V07 that rectify temporal artifacts caused by the time interpolation that fills the gaps between the various passive microwave (PMW) sensor overpasses. A second priority is to improve the homogeneity between the TRMM and GPM eras by characterizing differences between the two eras, determining the causes of these differences, and applying corrections as feasible, perhaps by enforcing spatial scale consistency (an overarching issue). Certainly, we must account for GPROF and the Combined Radar-Radiometer Algorithm converting to Machine Learning schemes in V08. Other priority topics include additional automated quality control for artifacts in the IR brightness temperatures and PMW precipitation fields, revisions to the specification algorithm for the probability of liquid precipitation, and accommodating new PMW sensors, which include the next generation of small-sats. We also consider the post-V08 landscape; the final GPM reprocessing will be restricted to fixing known code or algorithmic errors. Nonetheless, there are several data sources on the horizon to consider, including more small-sat PMW radiometers, AVHRR-based precipitation estimates (most useful in high latitudes), and the ISCCP-Next Generation and GEO-Ring projects that could provide easy access to multiple geosynchronous satellite channels and enable significantly improved algorithms compared to GEO-IR alone.

George J. Huffman↗

Verification, Validation, and Calibration Through a Causal Lens

While typical validation and verification approaches focus on identifying the associations between data elements using statistical and machine learning methods, the novel methods in this paper focus instead on identifying causal relationships between data elements. Statistical and machine-learning-based approaches are strictly data-driven, meaning that they provide quantitative comparison measures between data sets without explicitly considering the hypotheses behind them. This can lead to the erroneous conclusion that, if two data sets are close enough, the models that generated them are similar. In addition, when experimental and simulated data differ to an extent that fails to meet the acceptance criteria, calibration techniques are used to tweak simulation model parameters to reduce the gap between the two types of data. This produces the false expectation that a simulation model will match reality. The methods presented in this paper move away from these strictly data-driven methods for validation and calibration toward more robust, model-driven methods based on causal inference. Causal inference aims to identify the possible mechanisms that might have generated data. Thus, this analysis targets the prediction of the effects when one (or more) of the identified mechanisms are altered. There are many approaches to identify, quantify, and illustrate causal relationships. For the scope of this paper, directed graphs are employed as causal models. If the directed graph lacks cycles, it is known as a directed acyclic graph. A node in such a graph represents an observed data element while a directed edge connecting two nodes represents a causal relationship between two variables. The developed causal methods are designed to extract causal models from simulation models and experimental data. Causal models capture the causal relationships between data elements (e.g., simulated and experimental data). In this context, validation and verification are performed by comparing causal models. The proposed approach does not only inform system analysts on how a simulation model matches real-world data, but also identifies elements of the simulation model that should be revised when discrepancies between simulation and experimental data are observed. Through these causal methods, analysts can identify the portion of the model equation(s) that are behind an edge connecting two variables. Hence, once the structural differences between causal models have been determined, model calibration can occur by changing only those model parameters that impact the identified causal relationships.

97 MATHEMATICS AND COMPUTING↗

Leveraging Natural Language Processing and Generative Models in Molecular Chemistry: Property Prediction and Novel Compound Generation

The accurate prediction of molecular properties is important for the rational design and the advancement of green chemistry and sustainable materials research. However, the predictive power of traditional computational chemistry methods is limited due to computational restrictions. Here, in this study, we examine an alternative approach to the accurate prediction of properties of organic compounds: natural language processing (NLP)-based molecular embedding. Using viscosity, partition coefficient (log P), and enthalpy of vaporization as test properties through a survey of comprehensive datasets comprising 5695 data points for viscosity, 25 870 data points for log P, and 2296 data points for enthalpy of vaporization. These are important properties for the design of greener, safer, and sustainable chemical processes. Models were trained using NLP methods such as Mol2vec and fine-tuned ChemBERTa, and results were compared with traditional input featurization techniques such as Morgan fingerprints and quantum chemistry derived sigma profiles and DFT features. Among the various machine learning models, Mol2vec demonstrated superior predictive capabilities, achieving the highest correlation coefficient (R 2 = 0.945) and lowest RMSE (0.106 mPa s) for viscosity, as well as high accuracy for log P and enthalpy of vaporization predictions. These findings establish the Mol2vec featurization technique, graph-convolutional neural networks (GCNN), and fine-tuned ChemBERTa model as powerful tools for predictive modeling of organic compounds properties, offering a significant improvement over previously used featurization techniques and opening up strategies for very-high-throughput computational screening. Finally, we integrated ML models with hybrid language-model-based generative adversarial networks (LM-GAN) to generate novel molecular sequences with desirable properties for different research applications. The ability to computationally design solvents with lower viscosity, lower log P, and lower enthalpy of vaporization offers a data-driven route to accelerating the discovery of sustainable alternatives to traditionally toxic solvents.

ChemBERTa↗

Aurora Detection From Nighttime Lights for Earth and Space Science Applications

This research leverages data from the Day/Night Band (DNB) of the Visible Infrared Imaging Radiometer (VIIRS) instrument onboard the Suomi National Polar-orbiting Partnership (S-NPP) satellite. We demonstrate the value of mining the VIIRS DNB for aurora and describe our use of unsupervised machine learning to create a binary mask for aurora occurrence. This mask can be used to flag aurora-contaminated observations for NASA's nighttime lights products for Earth science applications. The identification of auroral regions can also be used for Space Weather applications, for example, for comparison with aurora forecast model and with other satellite- or ground-based aurora observations. The DNB is a broadband channel that is sensitive to wavelengths from 500 to 900 nm, which covers most of the visible light spectrum, and as the name implies, captures light even at night with a sensitivity at the nanowatt level. This band is suitable for aurora observations since the light emitted by the aurora tends to be dominated by emissions from atomic oxygen, resulting in a greenish glow at a wavelength of 557.7 nm, especially at an altitude of 110 km. This study compares the global nighttime derived aurora regions for 17 and 18 March with the NOAA Space Weather Prediction Center's (SWPC) probability product for the St. Patrick's Day geomagnetic storm in 2015. VIIRS sensors are slated to be added to the next generation of polar-orbiting operational satellites. Our novel automated approach to aurora identification opens up an efficient way to leverage this unique data source.

Aurora↗

Federated Machine Learning-Based Anomaly Detection System for Synchrophasor Network Using Heterogeneous Data Sets: Preprint

Synchrophasor technology is widely deployed in the energy management system to monitor the grid health at micro level and perform necessary corrective actions in real time; however, integrated phasor devices and data aggregators are exposed to several cybersecurity threats. This paper proposes a federated ML(FML)-based ADS to detect several data integrity attacks in the synchrophasor network. The proposed approach integrates the horizontal FML technique and consists of substation-based local models and a control center-based global model. The proposed methodology includes training local models using heterogeneous data sets that include network and grid information and updating the global model through multiple iterations by sharing model gradients. Finally, the trained global model is applied to identify cyberattacks, normal operation, and physical events. To validate the proof of concept, we used synthetic data sets generated by Mississippi State University and Oak Ridge National Laboratory for training and testing the classification models using the National Renewable Energy Laboratory's high performance computing resources. Our experimental results, computed through several performance measures, reveal that the proposed approach shows consistent performance during the binary, three-class, and multiclass classifications while ensuring privacy of synchrophasor data.

anomaly detection system↗

A Bayesian framework for adsorption energy prediction on bimetallic alloy catalysts

Abstract For high-throughput screening of materials for heterogeneous catalysis, scaling relations provides an efficient scheme to estimate the chemisorption energies of hydrogenated species. However, conditioning on a single descriptor ignores the model uncertainty and leads to suboptimal prediction of the chemisorption energy. In this article, we extend the single descriptor linear scaling relation to a multi-descriptor linear regression models to leverage the correlation between adsorption energy of any two pair of adsorbates. With a large dataset, we use Bayesian Information Criteria (BIC) as the model evidence to select the best linear regression model. Furthermore, Gaussian Process Regression (GPR) based on the meaningful convolution of physical properties of the metal-adsorbate complex can be used to predict the baseline residual of the selected model. This integrated Bayesian model selection and Gaussian process regression, dubbed as residual learning, can achieve performance comparable to standard DFT error (0.1 eV) for most adsorbate system. For sparse and small datasets, we propose an ad hoc Bayesian Model Averaging (BMA) approach to make a robust prediction. With this Bayesian framework, we significantly reduce the model uncertainty and improve the prediction accuracy. The possibilities of the framework for high-throughput catalytic materials exploration in a realistic setting is illustrated using large and small sets of both dense and sparse simulated dataset generated from a public database of bimetallic alloys available in Catalysis-Hub.org.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Automating the Analysis of Large Language Models Responses through Zero-Shot Question Answering

Recent advancements in Large Language Models (LLMs) have shown significant potential in various applications, yet their evaluation, particularly in zero-shot question answering scenarios, remains a challenging task. In this study, our objective was to explore precision metrics for Large Language Models (LLM) and design and implement a software pipeline to automatically evaluate LLMs' outputs under zero-shot question answering. Zero-shot question answering involves a model providing answers to questions about topics it hasn't seen during training. It leverages the principles of zero-shot learning by relying on semantic understanding and generalization from related knowledge. The data used was metadata from medical databases on congenital heart disease. We explored eleven LLM metrics and selected three for our evaluation: BLEU, BERTScore, and MoverScore. BLEU calculates a score based on the overlap of n-grams (contiguous sequences of n items, typically words) between the machine-generated translation and the reference translations. Higher BLEU scores indicate better correspondence between the machine-generated and human-generated translations. BERTScore is a metric used to evaluate the quality of machine-generated text by measuring the similarity of token embeddings produced by BERT (Bidirectional Encoder Representations from Transformers) between the generated text and reference text. MoverScore is a metric that quantifies the dissimilarity between the distributions of word embeddings from machine-generated text and reference text, emphasizing semantic similarity over exact token overlap. We also introduced HBKI, a composite metric summarizing these approaches. We tested five models —GPT-3, Llama-2, Gemini 1.5 Pro, Solar 10.7B, and Mixtral-8x7b. Our software pipeline, designed and implemented using Object-Oriented Programming principles, allows users to customize the selection and extraction of features for topics of interest in their own research. Our results show that MoverScore delivered the most precise evaluation of the LLM's outputs, while Mixtral-8x7b achieved the best overall performance in extracting metadata from the databases.

97 MATHEMATICS AND COMPUTING↗