Search NASA⌕ Search

SEARCH · Search NASA

Results for “Vectorized algorithm”

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 325 records · Page 18

Theoretical & Experimental Research in Weak, Electromagnetic & Strong Interactions

The conducted research spans a wide range of topics in the theoretical, experimental and phenomenological aspects of elementary particle interactions. Theory projects involve topics in both the energy frontier and the intensity frontier. The experimental research involves energy frontier with the ATLAS Collaboration at the Large Hadron Collider (LHC). In theoretical research, novel ideas going beyond the Standard Model with strong theoretical motivations were proposed, and their experimental tests at the LHC and forthcoming neutrino facilities were outlined. These efforts fall into the following broad categories: (i) TeV scale new physics models for LHC Run 2, including left-right symmetry and trinification symmetry, (ii) unification of elementary particles and forces, including the unification of gauge and Yukawa interactions, (iii) supersummetry and mechanisms of supersymmetry breaking, (iv) superworld without supersymmetry, (v) general models of extra dimensions, (vi) comparing signals of extra dimensions with those of supersymmetry, (vii) models with mirror quarks and mirror leptons at the TeV scale, (viii) models with singlet quarks and singlet Higgs and their implications for Higgs physics at the LHC, (ix) new models for the dark matter of the universe, (x) lepton flavor violation in Higgs decays, (xi) leptogenesis in radiative models of neutrino masses, (xii) light mediator models of non-standard neutrino interactions, (xiii) anomalous muon decay and short baseline neutrino anomalies, (xiv) baryogenesis linked to nucleon decay, and (xv) a new model for recently observed diboson resonance at the LHC and its other phenomenological implications. The experimental High Energy Physics group has been, and continues to be, a successful and productive contributor to the ATLAS experiment at the LHC. Members of the group performed search for gluinos decaying to stop and top quarks, new heavy gauge bosons decaying to top and bottom quarks, and vector-like quarks produced in pairs and decaying to light quarks. Members of the OSU group played a leading role in the detailed optimization studies for the future ATLAS Inner Tracker (ITk), which will be installed during the Phase-II upgrade, replacing the current tracking system. The proposed studies aim to enhance the ATLAS discovery potential in the high-luminosity LHC era. The group members have contributed to the calibration of algorithms for identifying boosted vector bosons and b-jets, which will help expand the ATLAS reach in many searches for new physics.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

3D reconstruction of the magnetic vector potential using model based iterative reconstruction

Lorentz transmission electron microscopy (TEM) observations of magnetic nanoparticles contain information on the magnetic and electrostatic potentials. Vector field electron tomography (VFET) can be used to reconstruct electromagnetic potentials of the nanoparticles from their corresponding LTEM images. The VFET approach is based on the conventional filtered back projection approach to tomographic reconstructions and the availability of an incomplete set of measurements due to experimental limitations means that the reconstructed vector fields exhibit significant artifacts. In this paper, we outline a model-based iterative reconstruction (MBIR) algorithm to reconstruct the magnetic vector potential of magnetic nanoparticles. We combine a forward model for image formation in TEM experiments with a prior model to formulate the tomographic problem as a maximum a-posteriori probability estimation problem (MAP). The MAP cost function is minimized iteratively to determine the vector potential. Here, a comparative reconstruction study of simulated as well as experimental data sets show that the MBIR approach yields quantifiably better reconstructions than the VFET approach.

36 MATERIALS SCIENCE↗

Optical Communication Link Atmospheric Attenuation Model

The Space Communications and Navigation (SCaN) Center for Networking, Integration, and Communications (SCENIC) user interface, which provides web accessible space mission simulation and communication system analysis capabilities using verified and validated analysis algorithms, can execute analyses including, but not limited to, line-of-sight, orbit propagation, and dynamic link budget calculations between sets of missions and/or assets. SCENIC's purpose is to provide NASA civil servants and contractors a user-friendly tool, integrated with model data, that can simulate and analyze a range of space mission architectures without the need for repeated and redundant modeling. Given the abundance and further future development of free space optical (FSO) communication channels within modern space infrastructure, the availability of a reliable optical link analysis capability is crucial for SCENIC users. The efforts outlined in this paper aim to provide a model for atmospheric attenuation of FSO communication links, both due to absorption/scattering and turbulence, to increase the accuracy of SCENIC's optical link assessment capabilities. A previous model existed for optical absorption/scattering within the SCaN Link Budget Tool, but it was not location specific for the Earth ground-based nodes, nor was the model optimized for run-time. The new model utilizes years of National Oceanic and Atmospheric Administration (NOAA) visibility data from ground station locations around the world. Visibility, along with the wavelength of the optical signal, are input parameters to calculate the optical specific attenuation, which is a parameter in the calculation of the slant-path attenuation. A final FSO atmospheric attenuation value is comprised of the absorption/scattering attenuation and the turbulence attenuation. A run-time efficient algorithm for the model was then developed and programmed in MATLAB ® . Due to the simple model and vectorization possible in MATLAB, the algorithm has an average run-time of less than one fourth of the run-time of the previous implementation.

Jack L Green↗

Simultaneous Mass Determination for Gravitationally Coupled Asteroids

The conventional least-squares asteroid mass determination algorithm allows us to solve for the mass of a large subject asteroid that is perturbing the trajectory of a smaller test asteroid. However, this algorithm is necessarily a first approximation, ignoring the possibility that the subject asteroid may itself be perturbed by the test asteroid, or that the encounter’s precise geometry may be entangled with encounters involving other asteroids. After reviewing the conventional algorithm, we use it to calculate the masses of 30 main-belt asteroids. Compared to our previous results, we find new mass estimates for eight asteroids (11 Parthenope, 27 Euterpe, 51 Neimausa, 76 Freia, 121 Hermione, 324 Bamberga, 476 Hedwig, and 532 Herculina) and significantly more precise estimates for six others (2 Pallas, 3 Juno, 4 Vesta, 9 Metis, 16 Psyche, and 88 Thisbe). However, we also find that the conventional algorithm yields questionable results in several gravitationally coupled cases. To address such cases, we describe a new algorithm that allows the epoch state vectors of the subject asteroids to be included as solve-for parameters, allowing for the simultaneous solution of the masses and epoch state vectors of multiple subject and test asteroids. We then apply this algorithm to the same 30 main-belt asteroids and conclude that mass determinations resulting from current and future high-precision astrometric sources (such as Gaia ) should conduct a thorough search for possible gravitational couplings and account for their effects.

79 ASTRONOMY AND ASTROPHYSICS↗

Assessment of Quantum ML Applicability for Climate Actions: Comparison of the Variational Quantum Classifier and the Quantum Support Vector Classifier with Classical ML Models

Climate change refers to significant and long-term alterations in the Earth’s climate patterns, typically resulting from human activities that increase greenhouse gas emissions. Addressing climate change is not merely an option but a necessity, demanding creative solutions and efforts from individuals, researchers, communities, and governments. Despite the capabilities of machine learning (ML) with data-driven solutions promising to combat climate change-related problems, they face challenges stemming from traditional computational methods and prolonged training times, impeding their practical utility. Recent strides in quantum computing have permeated diverse domains, spanning from manufacturing engineering and pharmaceutical discovery to the latest frontier of detecting climate anomalies. With the potential to substantially reduce time and computational complexity, quantum computing shows promise in addressing climate change impacts. Its distinctive features will enable the concurrent exploration of expansive solution spaces, making it well-suited for analyzing extensive climate datasets, simulating intricate climate models, optimizing resource allocation, and discerning patterns in climate data for mitigation and adaptation endeavors. This study explores the potential of using Quantum machine learning (QML) techniques on climate and weather data obtained from NASA Giovannis. We used two QML algorithms, the Quantum Support Vector Classifier (QSVC) and the Variational Quantum Classifier (VQC) models, using the IBM Qiskit ML 0.7.2 ecosystem. We used an actual 127-Qubit IBM Quantum Computer (IBM 127-qubit Eagle) in this study. The methodology and results sections describe the experiences gained from applying and evaluating quantum ML results on climate and weather data obtained from NASA satellites as a novel practical application of quantum computing.

Earth Observational Data↗

Image data compression with vector quantization in the transform domain

In this paper, an algorithm is presented for image data compression based upon vector quantization of the two-dimensional discrete cosine transformed coefficients. The ac energies of the transformed blocks are used to classify them into eight different ac classes. The ac coefficients of the transformed blocks of class one are set to zero, while those of classes two through eight are transmitted by seven different code books. The dc coefficients of all eight classes are scalar quantized by an adaptive uniform quantizer. As a result, only 4.5 bits instead of eight bits are required to transmit the dc coefficient with negligible additional degradation. Overall, this algorithm requires approximately 0.75 bits per pixel and gives an average reconstruction error of 7.1.

Abdelwahab, A. A.↗

The nondeterministic divide

The nondeterministic divide partitions a vector into two non-empty slices by allowing the point of division to be chosen nondeterministically. Support for high-level divide-and-conquer programming provided by the nondeterministic divide is investigated. A diva algorithm is a recursive divide-and-conquer sequential algorithm on one or more vectors of the same range, whose division point for a new pair of recursive calls is chosen nondeterministically before any computation is performed and whose recursive calls are made immediately after the choice of division point; also, access to vector components is only permitted during activations in which the vector parameters have unit length. The notion of diva algorithm is formulated precisely as a diva call, a restricted call on a sequential procedure. Diva calls are proven to be intimately related to associativity. Numerous applications of diva calls are given and strategies are described for translating a diva call into code for a variety of parallel computers. Thus diva algorithms separate logical correctness concerns from implementation concerns.

Charlesworth, Arthur↗

Data-driven cyber attack detection and mitigation for decentralized wide-area protection and control in smart grids

Modern power systems have already evolved into complicated cyber physical systems (CPS), often referred to as smart grids, due to the continuous expansion of the electrical infrastructure, the augmentation of the number of heterogeneous system components and players, and the consequential application of a diversity of information and telecommunication technologies to facilitate the Wide Area Monitoring, Protection and Control (WAMPAC) of the day-to-day power system operation. Because of the reliance on cyber technologies, WAMPAC, among other critical functions, is prone to various malicious cyber attacks. Successful cyber attacks, especially those sabotage the operation of Bulk Electric System (BES), can cause great financial losses and social panics. Application of conventional IT security solutions is indispensable, but it often turns out to be insufficient to mitigate sophisticated attacks that deploy zero-day vulnerabilities or social engineering tactics. To further improve the resilience of the operation of smart grids when facing cyber attacks, it is desirable to make the WAMPAC functions per se capable of detecting various anomalies automatically, carrying out adaptive activity adjustments in time and thus staying unimpaired even under attack. Most of the existing research efforts attempt to achieve this by adding novel functional modules, such as model-based anomaly detectors, to the legacy centralized WAMPAC functions. In contrast, this dissertation investigates the application of data-driven algorithms in cyber attack detection and mitigation within a decentralized architecture aiming at improving the situational awareness and self-adaptiveness of WAMPAC. First part of the research focuses on the decentralization of System Integrity Protection Scheme (SIPS) with Multi-Agent System (MAS), within which the data-driven anomaly detection and optimal adaptive load shedding are further explored. An algorithm named as Support Vector Machine embedded Layered Decision Tree (SVMLDT) is proposed for the anomaly detection, which provides satisfactory detection accuracy as well as decision-making interpretability. The adaptive load shedding is carried out by every agent individually with dynamic programming. The load shedding relies on the load profile propagation among peer agents and the attack adaptiveness is accomplished by maintaining the historical mean of load shedding proportion. Load shedding only takes place after the consensus pertaining to the anomaly detection is achieved among all interconnected agents and it serves the purpose of mitigating certain cyber attacks. The attack resilience of the decentralized SIPS is evaluated using IEEE 39 bus model. It is shown that, unlike the traditional centralized SIPS, the proposed solution is able to carry out the remedial actions under most Denial of Service (DoS) attacks. The second part investigates the clustering based anomalous behavior detection and peer-assisted mitigation for power system generation control. To reduce the dimensionality of the data, three metrics are designed to interpret the behavior conformity of generator within the same balancing area. Semi-supervised K-means clustering and a density sensitive clustering algorithm based on Hierarchical DBSCAN (HDBSCAN) are both applied in clustering in the 3D feature space. Aiming to mitigate the cyber attacks targeting the generation control commands, a peer-assisted strategy is proposed. When the control commands from control center is detected as anomalous, i.e. either missing or the payload of which have been manipulated, the generating unit utilizes the peer data to infer and estimate a new generation adjustment value as replacement. Linear regression is utilized to obtain the relation of control values received by different generating units, Moving Target Defense (MTD) is adopted during the peer selection and 1-dimensional clustering is performed with the inferred control values, which are followed by the final control value estimation. The mitigation strategy proposed requires that generating units can communicate with each other in a peer-to-peer manner. Evaluation results suggest the efficacy of the proposed solution in counteracting data availability and data integrity attacks targeting the generation controls. However, the strategy stays effective only if less than half of the generating units are compromised and it is not able to mitigate cyber attacks targeting the measurements involved in the generation control.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Attenuated Vector Tomography -- An Approach to Image Flow Vector Fields with Doppler Ultrasonic Imaging

The measurement of flow obtained using continuous wave Doppler ultrasound is formulated as a directional projection of a flow vector field. When a continuous ultrasound wave bounces against a flowing particle, a signal is backscattered. This signal obtains a Doppler frequency shift proportional to the speed of the particle along the ultrasound beam. This occurs for each particle along the beam, giving rise to a Doppler velocity spectrum. The first moment of the spectrum provides the directional projection of the flow along theultrasound beam. Signals reflected from points further away from the detector will have lower amplitude than signals reflected from points closer to the detector. The effect is very much akin to that modeled by the attenuated Radon transform in emission computed tomography.A least-squares method was adopted to reconstruct a 2D vector field from directional projection measurements. Attenuated projections of only the longitudinal projections of the vector field were simulated. The components of the vector field were reconstructed using the gradient algorithm to minimize a least-squares criterion. This result was compared with the reconstruction of longitudinal projections of the vector field without attenuation. Ifattenuation is known, the algorithm was able to accurately reconstruct both components of the full vector field from only one set of directional projection measurements. A better reconstruction was obtained with attenuation than without attenuation implying that attenuation provides important information for the reconstruction of flow vector fields.This confirms previous work where we showed that knowledge of the attenuation distribution helps in the reconstruction of MRI diffusion tensor fields from fewer than the required measurements. In the application of ultrasound the attenuation distribution is obtained with pulse wave transmission computed tomography and flow information is obtained with continuous wave Doppler.

60 APPLIED LIFE SCIENCES↗

Application of mixed kernels function (MKF) based support vector regression model (SVR) for CO 2 – Reservoir oil minimum miscibility pressure prediction

Carbon dioxide (CO 2 ) injection into oil reservoirs is considered a mature enhanced oil recovery (EOR) technique for conventional reservoirs. The local displacement efficiency of the CO 2 -EOR process is highly dependent on the minimum miscibility pressure (MMP), estimating this parameter is critical to design of the CO 2 injection process. Traditional empirical methods to test the CO 2 -oil MMP are time consuming and expensive; derived correlations are fast but not accurate. Therefore, an efficient and reliable method to determine MMP is beneficial. Here in this study, a mixed kernels function (MKF) based support vector regression (SVR) model was developed and used to predict the MMP for both pure and impure CO 2 injection cases. Four parameters were chosen as input parameters: (1) reservoir temperature; (2) average critical temperature; (3) molecular weight of pentane plus (C 5+ ) fraction of crude oil, and; (4) the ratio of volatile components to intermediate components in crude oil. MMP was selected as the desired output parameter to train and test this newly developed model. The performance of basic kernels function based SVR model is compared with that of this newly developed MKF-SVR model. The well-trained MFK-SVR was compared with three well-established published correlations, demonstrated the highest correlation coefficient (R of 0.9381), lowest root mean square error (RMSE of 1.9151), smallest average absolute error (AAE of 1.1406) and maximum absolute error (MAE of 4.6291). We believe that the proposed MFK-SVM model is a more reliable and stable regression model to predict MMP. In addition, a sensitivity analysis was conducted to evaluate the physical correctness. It indicates that the predicted results from the newly developed model are in excellent agreement with previous empirical work.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

QUASI-STELLAR OBJECT SELECTION ALGORITHM USING TIME VARIABILITY AND MACHINE LEARNING: SELECTION OF 1620 QUASI-STELLAR OBJECT CANDIDATES FROM MACHO LARGE MAGELLANIC CLOUD DATABASE

We present a new quasi-stellar object (QSO) selection algorithm using a Support Vector Machine, a supervised classification method, on a set of extracted time series features including period, amplitude, color, and autocorrelation value. We train a model that separates QSOs from variable stars, non-variable stars, and microlensing events using 58 known QSOs, 1629 variable stars, and 4288 non-variables in the MAssive Compact Halo Object (MACHO) database as a training set. To estimate the efficiency and the accuracy of the model, we perform a cross-validation test using the training set. The test shows that the model correctly identifies {approx}80% of known QSOs with a 25% false-positive rate. The majority of the false positives are Be stars. We applied the trained model to the MACHO Large Magellanic Cloud (LMC) data set, which consists of 40 million light curves, and found 1620 QSO candidates. During the selection none of the 33,242 known MACHO variables were misclassified as QSO candidates. In order to estimate the true false-positive rate, we crossmatched the candidates with astronomical catalogs including the Spitzer Surveying the Agents of a Galaxy's Evolution LMC catalog and a few X-ray catalogs. The results further suggest that the majority of the candidates, more than 70%, are QSOs.

79 ASTRONOMY AND ASTROPHYSICS↗

Vector-Matrix Multiplication Engine for Neuromorphic Computation with a CBRAM Crossbar Array [Slides]

The core function of many neural network algorithms is the dot product, or vector matrix multiply (VMM) operation. Crossbar arrays utilizing resistive memory elements can reduce computational energy in neural algorithms by up to five orders of magnitude compared to conventional CPUs. Moving data between a processor, SRAM, and DRAM dominates energy consumption. By utilizing analog operations to reduce data movement, resistive memory crossbars can enable processing of large amounts of data at lower energy than conventional memory architectures.

97 MATHEMATICS AND COMPUTING↗

Improved MPI collectives for MPI processes in shared address spaces

As the number of cores per node keeps increasing, it becomes increasingly important for MPI to leverage shared memory for intranode communication. This paper investigates the design and optimization of MPI collectives for clusters of NUMA nodes. We develop performance models for collective communication using shared memory and we demonstrate several algorithms for various collectives. Experiments are conducted on both Xeon X5650 and Opteron 6100 InfiniBand clusters. The measurements agree with the model and indicate that different algorithms dominate for short vectors and long vectors. We compare our shared-memory allreduce with several MPI implementations-Open MPI, MPICH2, and MVAPICH2-that utilize system shared memory to facilitate interprocess communication. On a 16-node Xeon cluster and 8-node Opteron cluster, our implementation achieves on geometric average 2.3X and 2.1X speedup over the best MPI implementation, respectively. Our techniques enable an efficient implementation of collective operations on future multi- and manycore systems.

Collective communication↗

Dosimetric Effect of Online Image-Guided Anatomical Interventions for Postprostatectomy Cancer Patients

Purpose: To assess daily variations in delivered doses in postprostatectomy patients, using kilovoltage cone-beam CT (CBCT) datasets acquired before and after interventions to correct for observed distortions in volume/shape of rectum and bladder. Methods and Materials: Seventeen consecutive patients treated with intensity-modulated radiotherapy to the prostate bed were studied. For patients with large anatomical variations, quantified by either a rectal wall displacement of >5 mm or bladder volume change of >50% on the CBCT compared with the planning CT, an intervention was performed to adjust the rectum and/or bladder filling. Cumulative doses over the pre- and post-intervention fractions were calculated by tracking the position of the planning CT voxels on different CBCTs using a deformable surface-mapping algorithm. Dose and displacements vectors were projected on two-dimensional maps, the minimal dose received by the highest 95% of the planing target volume (PTV D95) and the highest 10% of the rectum volume (D10) as well as the bladder volume receiving >2 Gy (V2) were evaluated. Results: Of 544 fractions, 96 required intervention. Median (range) number of interventions per patient was 5 (2-12). Compared with the planning values, the mean (SD) pre- vs. postintervention value for PTV D95 was -2% (2%) vs. -1% (2%) (p < 0.12), for rectum D10 was -1% (4%) vs. +1% (4%) (p < 0.24), and for bladder V2 was +6% vs. +20% (p < 0.84). Conclusions: Interventions to reduce treatment volume deformations due to bladder and rectum fillings are not necessary when patients receive daily accurate CBCT localization, and the frequency of those potential interventions is low. However, for hypofractionated treatments, the relative frequency can significantly increase, and interventions can become more dosimetrically beneficial.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Grid-based minimization at scale: Feldman-Cousins corrections for light sterile neutrino search

High Energy Physics (HEP) experiments generally employ sophisticated statistical methods to present results in searches of new physics. In the problem of searching for sterile neutrinos, likelihood ratio tests are applied to short-baseline neutrino oscillation experiments to construct confidence intervals for the parameters of interest. The test statistics of the form Δχ2 is often used to form the confidence intervals, however, this approach can lead to statistical inaccuracies due to the small signal rate in the region-of-interest. In this paper, we present a computational model for the computationally expensive Feldman-Cousins corrections to construct a statistically accurate confidence interval for neutrino oscillation analysis. The program performs a grid-based minimization over oscillation parameters and is written in C++. Our algorithms make use of vectorization through Eigen3, yielding a single-core speed-up of 350 compared to the original implementation, and achieve MPI data parallelism by employing DIY. We demonstrate the strong scaling of the application at High-Performance Computing (HPC) sites. We utilize HDF5 along with HighFive to write the results of the calculation to file.

Wospakrik, Marianette↗

AdenPredictor: accurate prediction of the adenylation domain specificity of nonribosomal peptide biosynthetic gene clusters in microbial genomes

Microbial natural products represent a major source of bioactive compounds for drug discovery. Among these molecules, nonribosomal peptides (NRPs) represent a diverse class that include antibiotics, immunosuppressants, anticancer agents, toxins, siderophores, pigments, and cytostatics. The discovery of novel NRPs remains a laborious process because many NRPs consist of nonstandard amino acids that are assembled by nonribosomal peptide synthetases (NRPSs). Adenylation domains (A-domains) in NRPSs are responsible for selection and activation of monomers appearing in NRPs. During the past decade, several support vector machine-based algorithms have been developed for predicting the specificity of the monomers present in NRPs. These algorithms utilize physiochemical features of the amino acids present in the A-domains of NRPSs. In this article, we benchmarked the performance of various machine learning algorithms and features for predicting specificities of NRPSs and we showed that the extra trees model paired with one-hot encoding features outperforms the existing approaches. Moreover, we show that unsupervised clustering of 453560 A-domains reveals many clusters that correspond to potentially novel amino acids. While it is challenging to predict the chemical structure of these amino acids, we developed novel techniques to predict their various properties, including polarity, hydrophobicity, charge, and presence of aromatic rings, carboxyl, and hydroxyl groups.

59 BASIC BIOLOGICAL SCIENCES↗

Adaptive variational ground state preparation for spin-1 models on qubit-based architectures

Here, we apply the adaptive variational quantum imaginary time evolution (AVQITE) method to prepare ground states of one-dimensional spin S = 1 models. We compare different spin-to-qubit encodings (standard binary, Gray, unary, and multiplet) with regard to the performance and quantum resource cost of the algorithm. Using state-vector simulations, we study two well-known spin-1 models: the Blume-Capel model of transverse-field Ising spins with single-ion anisotropy, and the XXZ model with single-ion anisotropy. We consider system sizes of up to 20 qubits, which corresponds to spin-1 chains up to length 10. We determine the dependence of the number of CNOT gates in the AVQITE state preparation circuit on the encoding, the initial state, and the choice of operator pool in the adaptive method. Independent of the choice of encoding, we find that the CNOT gate count scales cubically with the number of spins for the Blume-Capel model and quartically for the anisotropic XXZ model. However, the multiplet and Gray encodings present smaller prefactors in the scaling relations. These results provide useful insights for the implementation of AVQITE on quantum hardware.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Efficient graph representation framework for chemical molecule similarity tasks

Graph data has emerged in numerous scientific domains and machine learning techniques have been widely used for analysis and learning of diverse data for prediction and decision. Machine learning techniques can readily address complex problems by leveraging their structural information. But graphs cannot be directly used for existing machine learning algorithms unless encoded as vectors. The problem of efficient representation of graphs is a substantial challenge in graph machine learning. In this paper, we propose a novel two-stage framework for the representation of chemical molecule graphs based on the strengths of Graph Isomorphism Networks (GINs) and Siamese autoencoders. In the first stage, the GIN model is constructed and trained using the structural information of chemical molecule graphs. Node attributes, edge attributes, and edge indices are used as input data, while graph attributes are used as labels. The GIN model effectively captures the structural characteristics of graphs and can accurately predict graph attributes, i.e., molecular properties. It also generates Graph Embeddings, represented as vectors that encode the structural information of graphs. In the second stage, Graph Embedding vectors are further optimized for downstream similarity tasks while preserving the graph structural information. The Siamese autoencoder is constructed and trained, which reduces the dimensionality of the Graph Embedding vectors, while maximizing the preservation of structural information in the original high-dimensional vectors. The resulting low-dimensional Graph Embeddings can be effectively utilized for tasks such as approximate nearest neighbor search. The experimental results demonstrate the effectiveness of our proposed framework in accurately predicting graph similarity.

Ma, Jiaji↗