11 research outputs found
Meta Learning for Graph Neural Networks
Deep learning has enabled incredible advances in pattern recognition such as the fields of computer vision and natural language processing. One of the most successful areas of deep learning is Convolutional Neural Networks (CNNs). CNNs have helped improve performance on many difficult video and image understanding tasks but are restricted to dense gridded structures. Further, designing their architectures can be challenging, even for image classification problems. The recently introduced graph CNNs can work on both dense gridded structures as well as generic graphs. Graph CNNs have been performing at par with traditional CNNs on tasks such as point cloud classification and segmentation, protein classification and image classification, while reducing the complexity of the network.
Graph CNNs provide an extra challenge in designing architectures due to more complex weight and filter visualization of generic graphs. Designing neural network architectures, yielding optimal performance, is a laborious and rigorous process. Hyperparameter tuning is essential for achieving state of the art results using specific architectures. Using a rich suite of predefined mutations, evolutionary algorithms have had success in delivering a high-quality population from a low-quality starter population. This thesis research formulates the graph CNN architecture design as an evolutionary search problem to generate a high-quality population of graph CNN model architectures for classification tasks on benchmark datasets
Epitaxial Zintl aluminide SrAl4 grown on a LaAlO3 substrate
abstract: Zintl phases are a class of intermetallic materials that have simultaneously ionic and covalent bonding resulting from charge transfer between two different atomic species. We present a combined first principles and experimental study of Zintl-phase SrAl4, which is grown in thin film form on the perovskite oxide LaAlO3 using molecular beam epitaxy. The structural properties are investigated using reflection-high-energy electron diffraction, x-ray diffraction, and cross-section transmission electron microscopy, which reveal relaxed epitaxial island growth. Photoelectron spectroscopy measurements verify the Zintl-Klemm nature of the bonding in the material and are utilized to determine the band offset and the work function of SrAl4, while transport measurements confirm its metallic behavior. The experimentally observed properties are confirmed using density functional calculations.Copyright (2013) American Institute of Physics. This article may be downloaded for personal use only. Any other use requires prior permission of the author and the American Institute of Physics. along with the following message: The following article appeared in (Physical Review B, 88 (4)) and may be found at (http://journals.aps.org/prb/abstract/10.1103/PhysRevB.88.045314)
Epitaxy of polar semiconductor Co3O4 (110): Growth, structure, and characterization
abstract: The (110) plane of Co[subscript 3]O[subscript 4] spinel exhibits significantly higher rates of carbon monoxide conversion due to the presence of active Co[superscript 3+] species at the surface. However, experimental studies of Co[subscript 3]O[subscript 4] (110) surfaces and interfaces have been limited by the difficulties in growing high-quality films. We report thin (10–250 Å) Co[subscript 3]O[subscript 4] films grown by molecular beam epitaxy in the polar (110) direction on MgAl[subscript 2]O[subscript 4] substrates. Reflection high-energy electron diffraction, atomic force microscopy, x-ray diffraction, and transmission electron microscopy measurements attest to the high quality of the as-grown films. Furthermore, we investigate the electronic structure of this material by core level and valence band x-ray photoelectron spectroscopy, and first-principles density functional theory calculations. Ellipsometry reveals a direct band gap of 0.75 eV and other interband transitions at higher energies. A valence band offset of 3.2 eV is measured for the Co[subscript 3]O[subscript 4]/MgAl[subscript 2]O[subscript 4] heterostructure. Magnetic measurements show the signature of antiferromagnetic ordering at 49 K. FTIR ellipsometry finds three infrared-active phonons between 300 and 700 cm[superscript −1].Copyright 2014 American Institute of Physics. This article may be downloaded for personal use only. Any other use requires prior permission of the author and the American Institute of Physics. along with the following message: The following article appeared in 115, 24 (2014) and may be found at http://dx.doi.org/10.1063/1.488504
Web-Based Bibliographic Services Offered by Top World and Indian University Libraries
Bibliographical services are services related to the library collection and access to those collections, whether print or online. In modern libraries, these services are provided online through website. Many studies have been carried out to study application of web 2.0 tools and web-based services offered by the libraries. There is need to study the basic fundamental library services offered via the web or websites in modern libraries. In this paper author studied tops world and Indian libraries and prepared inventory of web-based bibliographic services offered by these libraries and analyzed the data in a quantitative way.</p
Characterization of Oxide Thin Films and Interfaces Using Transmission Electron Microscopy
abstract: Multifunctional oxide thin-films grown on silicon and several oxide substrates have been characterized using High Resolution (Scanning) Transmission Electron Microscopy (HRTEM), Energy-Dispersive X-ray Spectroscopy (EDX), and Electron Energy-Loss Spectroscopy (EELS). Oxide thin films grown on SrTiO3/Si pseudo-substrate showed the presence of amorphised SrTiO3 (STO) at the STO/Si interface. Oxide/oxide interfaces were observed to be atomically clean with very few defects.
Al-doped SrTiO3 thin films grown on Si were of high crystalline quality. The Ti/O ratio estimated from EELS line scans revealed that substitution of Ti by Al created associated O vacancies. The strength of the crystal field in STO was measured using EELS, and decreased by ~1.0 eV as Ti4+ was substituted by Al3+. The damping of O-K EELS peaks confirmed the rise in oxygen vacancies. For Co-substituted STO films grown on Si, the EDS and EELS spectra across samples showed Co doping was quite random. The substitution of Ti4+ with Co3+ or Co2+ created associated oxygen vacancies for charge balance. Presence of oxygen vacancies was also confirmed by shift of Ti-L EELS peaks towards lower energy by ~0.4 eV. The crystal-field strength decreased by ~0.6 eV as Ti4+ was partially substituted by Co3+ or Co2+.
Spinel Co3O4 thin films grown on MgAl2O4 (110) were observed to have excellent crystalline quality. The structure of the Co3O4/MgAl2O4 interface was determined using HRTEM and image simulations. It was found that MgAl2O4 substrate is terminated with Al and oxygen. Stacking faults and associated strain fields in spinel Co3O4 were found along [111], [001], and [113] using Geometrical Phase Analysis.
NbO2 films on STO (111) were observed to be tetragonal with lattice parameter of 13.8 Å and NbO films on LSAT (111) were observed to be cubic with lattice parameter of 4.26 Å. HRTEM showed formation of high quality NbOx films and excellent coherent interface. HRTEM of SrAl4 on LAO (001) confirmed an island growth mode. The SrAl4 islands were highly crystalline with excellent epitaxial registry with LAO. By comparing HRTEM images with image simulations, the interface structure was determined to consist of Sr-terminated SrAl4 (001) on AlO2-terminated LAO (001).Dissertation/ThesisDoctoral Dissertation Physics 201
Guest Editorial: Special Issue on Data Analytics and Machine Learning for Network and Service Management - Part II
Data-Intensive System
Oculomics meets exposomics:a roadmap for applying multi-modal ocular biomarkers in precision environmental health research
Precision environmental health (PEH) is an emerging field that seeks to understand how diverse environmental exposures interact with individual biological and genetic factors to influence health outcomes. While recent advances in exposomics have enabled systematic characterization of the exposome, the integrated compilation of all physical, chemical, biological, and psychosocial influences that affect biology and health, identifying and developing sensitive biomarkers remains a critical challenge. The human eye offers unique potential for non-invasive biomarker discovery. Ocular biomarkers can be utilized not only for diagnostics and therapeutic responses of ocular diseases, but also for monitoring environmental exposures and predicting systemic health outcomes. Retinal imaging modalities such as color fundus photography, optical coherence tomography, and optical coherence tomography angiography capture biomarkers linked to environmental exposures and systemic conditions like cardiovascular and neurodegenerative diseases, a field known as oculomics. Similarly, ocular fluids, such as tears, aqueous humor and vitreous humor, may also reflect pollution-induced oxidative stress and inflammation and systemic health conditions. This paper summarizes current evidence on how ocular biomarkers can bridge environmental exposures and systemic health outcomes, and proposes future research directions using state of the art methodologies such as exposome-wide association studies, high dimensional mediation analysis, and multi-modal foundation models. Despite encouraging progress, significant challenges remain, including the need for large and standardized datasets, rigorous validation, and ethical safeguards to ensure equitable application. Advances in artificial intelligence, including federated learning, alongside global consortium efforts, will be essential to overcome these barriers. Addressing these gaps will unlock the full potential of oculomics and exposomics, advancing the goals of precision environmental health
