56016 research outputs found
Sort by
FunRank : Finding 1-day Vulnerabilities with Data-flow Analysis
No full textDepartment of Computer Science and Engineeringclos
?????? ???????????? ???????????? ????????? AumoML??? ?????? ?????? ????????????
No full textDepartment of Industrial EngineeringIn recent years, AutoML has emerged as a promising technique for reducing computational and time cost by automating the development of machine learning models. Existing AutoML tools cannot be applied directly to process predictive monitoring (PPM), because they do not support several configuration param- eters that are PPM-specific, such as trace bucketing or encoding. In other words, they are only specialized in finding the best configuration of machine learning model hyperparameters. In this thesis, we present a simple yet extensible framework for AutoML in PPM. The framework uses genetic algorithms to explore a configuration space containing both PPM-specific parameters and the traditional machine learning model hyperparameters. We design four different types of experiments to verify the effectiveness of the proposed approach, comparing its performance in respect of random search of the configuration space, using two pub- licly available event logs. The results demonstrate that the proposed approach outperforms consistently the random search.ope
Improving the Performance of Big Data Analytics Platforms by Task and I/O Granularity Adjustment
Full text linkDepartment of Computer Science and EngineeringWith the massive increase in the amount of semi-structured and unstructured web data, big data analytics platforms have emerged and started to evolve rapidly. Apache Hadoop has been developed for batch processing on a large dataset, and systems for interactive and general purpose applications have been developed alongside NoSQL databases. Numerous efforts have been made to improve the performance of Hadoop and NoSQL databases, including utilizing a new device called NVMM for NoSQL databases. Nonetheless, their performance is still far from satisfactory due to inadequate granularity for tasks and I/O. In this dissertation, we present novel techniques to improve the performance of Apache Hadoop and NVMM-based LSM-tree by adjusting task and I/O granularity.
First, we analyze YARN container overhead and present dynamic input split size adjustment scheme, which can logically combine multiple HDFS blocks and increase the input size of each container, thereby enabling a single map wave and reducing the number of containers and their initialization overhead. Experimental results shows that we can avoid recurring container overhead by selecting the right size for input splits and reducing the number of containers. Second, we present a novel HDFS block coalescing scheme that mitigates the YARN con tainer overhead. Our assorted block coalescing scheme combines multiple HDFS blocks and creates large input splits of various sizes, reducing the number of containers and their initializa tion overhead. Our experimental study shows the block coalescing scheme significantly reduces the container overhead while it achieves good load balancing and job scheduling fairness without impairing the degree of overlap between map phase and reduce phase.
Third, we discuss design choice of using NVMM for indexing structure in NoSQL databases and present ZipperDB, a key-value store that redesigns LSM-tree for byte-addressable persistent memory. To benefit from the byte-addressability of persistent memory, ZipperDB employs byte addressable persistent SkipLists and performs Zipper Compaction, a novel in-place compaction algorithm that merges two adjacent persistent SkipLists without compromising the failure atomicity. The byte-addressable compaction helps mitigate the write amplification problem, which is known to be the root cause of the write stall problem in LSM-tree.
Finally, we present ListDB, a write-optimized key-value store for NVMM to overcome the gap between DRAM and NVMM write latencies and thereby, resolve the write stall problem. ListDB consists of three novel techniques: (i) byte-addressable Index-Unified Logging, which incrementally converts write-ahead logs into SkipLists, (ii) Braided SkipList, a simple NUMA aware SkipList that effectively reduces the NUMA effects of NVMM, and (iii) NUMA-aware Zipper Compaction. Using the three techniques, ListDB makes background flush and com paction fast enough to resolve the infamous write stall problem and shows 1.6x and 25x higher write throughputs than PACTree and Intel Pmem-RocksDB, respectively.ope
Magnitude- and order-controlling of frequency modulation with nonlinear acoustic metamaterials.
Full text linkDepartment of Mechanical EngineeringThis thesis aims to control the magnitude and order of frequency modulation with nonlinear acoustic metamaterials. Frequency modulation is a nonlinear phenomenon that changes the frequency of a wave. Since frequency modulation can easily and actively provide waves with the desired frequency, it has been widely used. However, the frequency modulation has a technical issue that the magnitude is small and the order cannot be changed because the nonlinearity of conventional medium is generally small and constant. To overcome this problem, we investigate the underlying physics of frequency modulation and propose metamaterials that can control the magnitude and order of frequency modulation. Based on the underlying physics, the magnitude and order of the frequency modulation can be controlled by effective material properties and nonlinear parameters, respectively. Therefore, the metamaterial for magnitude-controlling of frequency modulation is designed based on the concept of coiling-up space, which can control effective material properties by the degree of coiling. On the other hand, the metamaterial for order-controlling of frequency modulation is designed based on the chevron beam, which can control nonlinear parameters by the initial angle. The performance of metamaterials for magnitude- and order-controlling of the frequency modulation are numerically validated.ope
2?????? ????????? ????????? MXene??? ????????????, ????????? ????????? ?????? ???????????? ??????
No full textDepartment of Materials Science and EngineeringExfoliated flakes of two-dimensional (2D) materials have achieved considerable step forward for industrial uses, including electronic devices, catalysts, and energy transformation and reservoir, due to their distinct qualities at atomic-level thicknesses. 2D layered solids possess a high surface-to-volume ratio and physical and chemical versatilities for operation as high-performance conducting materials in electrochemical energy systems, which make them distinctive from their bulk counterparts. Graphene and its analogueshave shown promise as the conducting materialshowever, their low structural stability and restricted inter-layer space severely limit their capacitance and rate performance. In addition, to achieve excellent cycling life and energy density, affordable nanomaterial-based electrode providing reversible redox reactions and mechanical stiffness should be developed. As materials for electrochemically operational conductors, MXenes, a category of 2D transition metal carbides, carbonitrides, and nitrides, have received substantial research interest. The 2D layered claylike structure of MXenes possessing an uncommon blend of ceramic and metallic qualities allows fast redox activity by fertile surface terminations while maintaining high conductivity. Furthermore, MXenes have significantly greater inter-layer spacing than carbon-based conductors, which facilitates rapid ion diffusion and volume expansion. Emphasis is given to processing routes for building conducting connects and mechanical scaffolds that provide easy access to electrolyte ions (associated with charge transfer) and increased capacities at high power rates for numerous cycles. For the realistic operation of MXenes for durable electrochemical catalysts and energy storage, preparation and stabilizing strategies on a large scale are also essential.
In this thesis, the conditions for synthesis and exfoliation processes were optimized to address the intricate quality control of 2D MXene products from the parent MAX phases, which is the most critical issue for the final attainments of devices or membranes. High-quality interfaces and well-crystalline nature of the as-exfoliated MXene flakes were confirmed by development of realistic and scalable multifunctional coating processes. At first, we demonstrate that a pelletization approach could produce a chemically ideal Ti3AlC2 MAX phase with excellent yields and manufacturability. (Chapter 2). The Ti3AlC2 was capable of being exfoliated into 1~2-layers-thick 2D Ti3C2Tx MXenes, and their adaptability was validated by the creation of a workable MXene ink that demonstrated coating-thickness uniformity and outstanding electrical performances as those of a film generated by vacuum filtering.
To suggest a scalable assembly technique for functional multiscale architecture, we report the fabrication of strong and conductingfilms using 2D MXenes as building blocks with guest species (Chapter 3). Ultrathin and robust B4C composite films for neutron shielding were fabricated without heating or casting, by mixing Ti3C2Tx MXene flakes. The evaluation of the layered structure of MXene flakes with different surface properties, contents, and sizes of the B4C filler enabled the optimization of the matrix capacity, resulting in the tunable neutron-shielding ability. The hybrid coatings exhibited excellent protection efficiency with mechanical flexibility that was promising for wearability.
In Chapter 4, a solution-processable production of patterned Ti3C2Tx MXene mesh via scalable blade coating was conducted to fabricate transparent devices with well-aligned assembly and excellent optoelectronic properties. On the other hand, a patterned 2D MXene electrode having a thickness less than 20 nm has been obtained by controlling the reactive ion etching conditions. Both macroscale MXene mesh and microscale 2D MXene conductors exhibited high electrical conductivity due to the percolation connect based on Ti3C2Tx MXene flakes and could be transferred, forming integration with other electronic materials (e.g., metals and channel materials) and soft polymers.
The final section reported the development of MXene adsorbent for precious metals with enhanced dispersibility to tackle the restacking problem of the delaminated flakes and severe oxidation issues in aqueous solvents (Chapter 5). In-plane crystallite dimensions and thickness of the dispersed 2D flakes of Ti3C2Tx MXene were controlled by a ball-milling process of the powder to increase the operational sites for adsorption and access to aqueous electrolytes. Based on the dispersibility, high surface area, and well preserved electrochemically operational sites, the developed MXene adsorbent exhibited excellent adsorption efficiency for Au and Ag througha chemisorption mechanism, which was described by empirical isotherms and dynamics.ope
Low-temperature crystallization of LaFeO3 perovskite with inherent catalytically surface for the enhanced oxygen evolution reaction
No full textThis study reports a facile and economic method for LaFeO3 perovskite crystallization process at low temperature range from 300 ??C to 500 ??C and an outstanding oxygen evolution reaction (OER) catalyst based on inherent catalytically surface. As a key material for low temperature synthesis, cyanogel-peroxo-complex as a metastable molecular precursor was synthesized via ligand exchange using potassium cyanide solution to provide superoxo (O2???) ligand to Fe-CN-La gel structure, leading to a high degree of crystallinity with ideal ABO3 stoichiometry at low temperatures (400 ??? 500 ??C). Electrocatalysts based on LaFeO3 nanoparticles were fabricated, showing an outstanding OER performance with low overpotential of ??? 438 mV at 100 mA/cm2 and small Tafel slope of 61 mV??dec???1 under alkaline conditions, better than commercialized available IrOx/C catalysts. Its OER performance is attributed to the inherent oxygen-deficient layer at the surface created at low temperature (300 ??C). Long-term stability test shows no significant change (< 1%) in the potential during 50 h, indicating a high stability of such catalysts
Terahertz virus-sized gold nanogap sensor
Full text linkWe demonstrated an ultra-sensitive terahertz virus detection method combined with virus-sized gold nanogaps filled with Al2O3. Large-area high-density 20 nm-gap rectangular loop structures, containing a resonant frequency in the terahertz range, were fabricated on a 4-inch wafer using atomic layer lithography. When target viruses with a 60 nm diameter were located on the nanogaps, we observed a significant redshift of the resonant peak already with an average number of about 100 viruses per unit loop due to the strong field confinement and enhancement near the gap. Furthermore, when the virus was tightly attached to an etched gap like a bridge connecting metals, its sensitivity is doubled compared to the unetched gap, which resulted in 400% more resonance frequency shift per single virus particle than our previous work. Full-wave simulations and theoretical calculations based on modal expansions were in good agreement with the experiments, revealing that the resonant transmission spectrum was mostly determined by the change in refractive index in a two-dimensional-like optical hotspot near the nanogap. A further step could be taken to increase sensitivity by tuning nanogap-loops to the absorption frequencies associated with the intermolecular vibrational modes of the viruses and fingerprinting them as well
Preparation of hexene-functionalized graphitic nanoplatelets for effective interaction with Nylon 6
No full textHexene-functionalized graphitic nanoplatelets (He-f-GN) were easily prepared using a mechanochemical reaction between solid graphite and liquid 1-hexene. The He-f-GN exhibited outstanding properties (e.g., high specific surface area, high crystallinity and so on) and could be well distributed in various solvents including formic acid. The He-f-GN/Nylon 6_X nanocomposites were simply prepared using the solution method, and showed excellent mechanical properties and thermal stability compared with the neat Nylon 6. Specifically, the tensile strength and Young's modulus of the He-f-GN/Nylon 6_1 nanocomposites increased by approximately 32.5% and 33.7%, respectively, compared to the neat Nylon 6 due to the special properties of the He-f-GN and its excellent compatibility with Nylon 6 chains. The He-f-GN also acts as nucleation sites, increasing the crystallinity of Nylon 6, and generated hydrogen bonds with the amide groups of the Nylon 6. The new filler, He-f-GN, provides an effective way to increase the performance of polymer, demonstrating good application prospects
Hybrid solid mesh structure for electron beam melting customized implant to treat bone cancer
No full textBone replacement implants manufactured by electron beam melting have been widely studied for use in bone tumor treatment. In this application, a hybrid structure implant with a combination of solid and lattice structures guarantees strong adhesion between bone and soft tissues. This hybrid implant must exhibit adequate mechanical performance so as to satisfy the safety criteria considering repeated weight loading during the patient???s lifetime. With a low volume of a clinical case, various shape and volume combinations, including both solid and lattice structures, should be evaluated to provide guidelines for implant design. This study examined the mechanical performance of the hybrid lattice by investigating two shapes of the hybrid implant and volume fractions of the solid and lattice structures, along with microstructural, mechanical, and computational analyses. These results demonstrate how hybrid implants may be designed to improve clinical outcomes by using patient-specific orthopedic implants with optimized volume fraction of the lattice structure, allowing for effective enhancement of mechanical performance as well as optimized design for bone cell ingrowth