Three Topics in Harmonic Analysis: Maximal Functions on Heisenberg Groups, Cotlar-Type Theorems and Wavelets on Sobolev Spaces

Srivastava, Rajula ProQuest Dissertations & Theses The University of 2022 해외박사(DDOD)

This thesis contains the results of the author from Roos (2021) (joint work with Joris Roos and Andreas Seeger), Rskor (2022), Srivastava (2020) and Srivastava (2020).In Chapter 2, we prove Lp → Lq estimates for local maximal operators associated with dilates of codimension two spheres in Heisenberg groups; these are sharp up to two endpoints. The results can be applied to improve bounds on sparse domination for global maximal operators. We also consider lacunary variants, and extensions to Metivier groups.In Chapter 3, we prove Lp → Lq estimates for the local maximal operator associated with dilates of the Koranyi sphere in Heisenberg groups. These estimates are sharp up to endpoints and imply new bounds on sparse domination for the corresponding global maximal operator. We also prove sharp Lp → Lq estimates for spherical means over the Koranyi sphere, which can be used to improve the sparse domination bounds in Ganguly (2021) for the associated lacunary maximal operator.In Chapter 4, we exhibit the necessary range for which functions in the Sobolev spaces Lsp can be represented as an unconditional sum of orthonormal spline wavelet systems, such as the Battle-Lemarie wavelets. We also consider the natural extensions to Triebel-Lizorkin spaces. This builds upon, and is a generalization of, previous work of Seeger and Ullrich (Seeger, 2017), where analogous results were established for the Haar wavelet system.In Chapter 5, we prove the Lp boundedness of a Cotlar-type maximal operator associated with a dyadic frequency decomposition of a Fourier multiplier, under a weak regularity assumption.

Corrosion of metals by soils

Srivastava, Kumkum Department of Chemistry, University of Allahabad 1981 국내박사

Steel Industry of India and Its Future Prospects

Srivastava, Abhishek Graduate school of international studies, Korea un 2009 국내석사

Content-Centric Smart Home Networking for Centralized Device Management

VARTIKA SRIVASTAVA 아주대학교 아주대학교 일반대학원 2016 국내석사

Smart home solutions provide home monitoring and automation, energy, health and media management services. However, smart home solutions are still not a major market product. This is because smart home devices are heterogeneous and so they face interoperability issues. Also, the data is identified by device address and so the user needs to know it in order to access the data. Secure data exchange is another major concern. The current solutions thus face issues of difficult setup, cumbersome maintenance and security lapses. To address these problems we propose a smart home solution based on Content Centric Networks (CCN). Name-based data access in CCN helps provide easier setup and maintenance of a smart home. CCN also helps in unifying heterogeneous devices at the network layer. Our scheme proposes the creation of virtual groups in which the devices can be controlled securely, and can be added flexibly regardless of physical locations. The management of devices and controllers is done centrally whereas all other communication takes place via a p2p connection. This provides better data synchronization, robust security and reduced device energy consumption. To validate our scheme, we implemented a proof of concept and measured time taken and packet overhead for setup and regular interactions. We found that our scheme outperforms existing work for both these metrics.

Enhancing Mathematical Word Problem Solving with Numerical Masking Strategy in Pre-trained Encoder-Decoder Language Models

Nilesh Kumar Srivastava 과학기술연합대학원대학교 한국과학기술정보연구원(KISTI) 2024 국내석사

Enhancing Mathematical Word Problem Solving with Numerical Masking Strategy in Pre-trained Encoder-Decoder Language Models Addressing Mathematical Word Problems (MWPs) within the domain of Natural Language Processing (NLP) is a challenging task. While decoder-based Large Language Models (LLMs) have demonstrated potential in MWP solving, their substantial size can be a hindrance. MWPs demand a unique fusion of language understanding and generation, making encoder-decoder models an enticing solution. However, these models grapple with specific complexities, especially the intricate handling of numerical values and mathematical relationships entwined within the context. In response, this study introduces a novel masking approach explicitly tailored for MWP solving, departing from the limitations of random masking. This approach trains the encoder-decoder framework to better predict essential numerical components and mathematical relationships within MWPs. This research endeavors to comprehensively explore the efficacy of the proposed masking approach in comparison to the conventional random masking methodology. Additionally, the study delves into the nuanced effects of varying masking percentages and span lengths during continual training on the performance of MWPs. By conducting a series of extensive experiments, this investigation un- ravels valuable insights, shedding light on the optimal configurations that contribute to an enhanced mathematical reasoning capacity within the realm of MWPs. The outcomes of this research present a promising narrative, showcasing substantial en- hancements in performance across a diverse range of datasets, including GSM8K, SVAMP, and MultiArith. Particularly noteworthy is the remarkable surge in accu- racy observed in the SVAMP dataset, soaring from an initial 12.90% to an impres- sive 43%. Similarly, the MultiArith dataset demonstrates a significant performance boost, with accuracy ascending from 26.88% to an elevated 49.53%. These findings underscore the effectiveness of our innovative masking approach in elevating the mathematical proficiency of encoder-decoder models dedicated to MWP solving. This study not only establishes the superiority of the proposed masking ap- proach but also provides valuable insights into the intricate dynamics of continual training, the impact of loss functions, and their combined influence on the pro- ficiency of encoder-decoder models in MWP solving. The nuanced variations ob- served in the results of different loss function combinations add depth to the research, contributing to the broader comprehension of optimizing models for MWP-solving tasks. Keywords: Mathematical Word Problems (MWPs); Language Models (LM); Masking Numerical Tokens: Random Masking; Mathematical Reasoning; Encoder- Decoder Models; Continual Training. 수치마스킹전략을이용한사전학습인코더-디코더언어모델의 수학문제풀이성능개선 자연어처리(NLP)태스크에서수학문제풀이태스크를(MWP)를해결하는 것은 어려운 태스크이다. 디코더 기반 LLM(거대 언어 모델)은 MWP 해결의 잠 재력을 입증했지만 이에 필요한 모델의 크기가 상당한 수준이어야 한다는 것이 한계이다. MWP는언어이해와생성의고유한융합을요구하므로인코더-디코더 모델이 대안이 될 수 있다. 그러나 이러한 모델은 특정 복잡성, 특히 컨텍스트 내에 얽혀 있는 수치 값 및 수학적 관계를 학습하여 추론하는 능력을 필요로 한 다. 이를 효과적으로 해결하기 위해, 이 연구에서는 무작위 마스킹의 한계에서 벗어나 MWP 해결을 위한 명시적으로 맞춤화된 새로운 마스킹 접근방식을 소 개하고 적용한다. 이 접근방식은 인코더-디코더 프레임워크가 MWP 내의 필수 수치구성요소와수학적관계를더잘예측하도록설계되었다. 본 연구에서는 기존의 무작위 마스킹 방법과 비교하여 제안된 마스킹 접 근방식의효율성을포괄적으로탐색하고결과를제시한다.또한이연구에서는 MWP성능에대한지속적인학습중다양한마스킹비율과스팬길이의미묘한 효과를조사하였다.일련의광범위한실험을수행함으로써이연구는귀중한통 찰력을밝혀내고MWP영역내에서향상된수학적추론능력에기여하는최적의 구성을 제안한다. 이 연구는 GSM8K, SVAMP 및 MultiArith를 포함한 다양한 NWP 데이터셋을 사용한 실험에서 성능이 크게 향상되었음을 보인다. 특히 주 목할 만한 점은 SVAMP 데이터셋에서 관찰된 정확도의 놀라운 상승으로, 초기 12.90%에서 43%까지성능향상을보였다.마찬가지로 MultiArith데이터셋에서 정확도가 26.88%에서 49.53%로증가함을보였다.이러한결과는MWP해결전 용인코더-디코더모델의수학적추론성능을높이는데있어제안한마스킹접근 방식이효과적이라는것을증명하였다. 이 연구는 제안된 마스킹 접근 방식의 우수성을입증할 뿐만 아니라 지속 적인 훈련의 복잡한 역학, 손실 함수의 영향 및 MWP 해결에서 인코더-디코더 모델의숙련도에대한결합된영향에대한통찰력을제공하였다.다양한손실함 수조합의결과에서관찰된미묘한변화는연구에깊이를더해MWP해결을위한 모델최적화에기여하였다. Keywords:수학단어예측(NWP);언어모델(LM);숫자토큰마스킹:무작위 마스킹;수학적추론;인코더-디코더모델;지속적인훈련.

Hydrometallurgical Recovery of Rhenium from Engine Reverted Turbine-Blade Superalloy

Rajiv Ranjan Srivastava University of Science and Technology 2017 국내박사

A hydrometallurgical recycling of the engine reverted turbine-blade superalloy has been investigated for the recovery of rare, strategic and critical metals therein. For this, a modern age nickel-based single crystal superalloy, CMSX-4 was used in the study. Present study emphasized on the particular recovery of rhenium and its separation from molybdenum, which is costlier than the other constituted elements in superalloy. The investigated process deliberates a two step leaching followed by two step solvent extraction and a subsequent crystallization operation. For a better understanding of entire process, the aqueous and complexation chemistry of rhenium was also studied in-depth and discussed in the thesis. An efficient leaching of superalloy was performed in a 2-step HCl leaching. In first step of HCl leaching, the maximum of thermal-barrier elements (viz. Al, Cr, W etc.) alloyed in superalloy were leached along with the base-alloyed elements (viz. Ni and Co). At an optimized leaching condition (lixiviant concentration = 4 mol∙L−1 HCl; solid-to-liquid ratio, S/L = 20 g∙L−1; temperature = 90 °C; time = 120 min), approximately 99.7% Ni, 97.5% Al, 96.9% Co, 99.2% Cr, 89.2% Ti and 46.6% Mo were leached. On that condition, only a 1.4% Re was leached. The tungsten precipitated as tungstic acid during HCl leaching was separated by an alkali-wash treatment, yielding > 98% W. Thus obtained residue was subjected to second leaching in HCl solution, where electro-generated chlorine was supplied as a source of oxidant. The electro-generated chlorine leaching was particularly chosen to get the rhenium into aqueous solution. At higher acid concentration (8 mol∙L−1 HCl) rhenium leaching was unaffected with Cl2 supply rate; whereas, on that concentration the leaching kinetics was also slow and after a prolonged leaching of 90 min the overall extraction was similar for 4–8 mol∙L−1 HCl. A shift of surface-diffusion (< 60 °C) to chemically-controlled (> 60 °C) reaction zone was clearly observed with a variance in leaching temperature from 40–80 °C. For the recovery of rhenium and its separation from molybdenum, two different techniques were adopted. Adsorption of rhenium onto activated carbon was investigated to obtain the maximum separation factor (βRe/Mo) by processing the first-step leach liquor. For this, the design of experiments (DOE) was done by Taguchi model of statistical design and the maximum βRe/Mo = 866 could be achieved when 70 mg∙L−1 Re and 250 mg∙L−1 Mo in solution of pH 5.0 was contacted with activated carbon at S/L = 18, temperature 10 °C for 30 min. Rhenium adsorbed on activated carbon was recovered by simply elution with ammonia solution. Whereas, the rhenium bearing leach liquor of second-step leaching was processed via applying the liquid-liquid separation techniques. The phosphinic acid extractant (Cyanex 272) was employed for a selective separation of the cationic MoO22+ from the leach liquor over neutral species of rhenium with a high separation factor (βMo/Re = 839; using 0.067 mol∙L−1 extractant). The molybdenum from the loaded organic phase was recovered by stripping with NaOH solution. Subsequently, rhenium was extracted from Mo-depleted aqueous solution by maiden use of a mixed phosphine oxides organic extractant (Cyanex 923). An investigation of parametric influences on rhenium extraction with Cyanex 923 exhibited a remarkable increase in the extraction efficiency with an increase in the HCl concentration. The extraction of rhenium increased from 9% to 98% with an increase in extractant concentration from 0.005 to 0.08 mol∙L−1 into the organic phase. Based on the McCabe–Thiele diagram, a simulation showed that 99.2% extraction efficiency could be attained in a two-stage counter-current extraction with an organic-to-aqueous (O/A) phase ratio of 2:3. The back-extraction of rhenium from loaded organic phase (containing 1.9 g∙L−1 Re) was achieved by contacting with NH4OH solution. After a three stages of counter-current stripping at an O/A = 5:1, an enriched solution of 9.45 g∙L−1 Re could obtain, yielding a 99.97% pure crystals of NH4ReO4 after the vacuum crystallization of Re-bearing solution.

Type I Interferons Negatively Regulate Foxp3+ Regulatory T Cells during Acute Viral Infection and Autoimmunity

Srivastava, Shivani University of Washington 2014 해외박사(DDOD)

CD4+ regulatory T (Treg) cells expressing the transcription factor Foxp3 are potent anti- inflammatory cells capable of restraining immune responses to both self- and foreign-antigens. In addition to preventing autoimmunity and immunopathology, Treg cells can also inhibit immune responses during viral, bacterial, and parasitic infections. Paradoxically, signals known to drive Treg proliferation, such as IL-2 and activated dendritic cells, are also abundant during infection when Treg activity may need to be curbed. How Treg activity is restrained during infection to allow for the generation of effective immune responses remains largely unclear. Type I interferons (IFNs) are a family of cytokines that coordinately regulate many cell types during viral infection, but their effects on Treg cells remain largely unknown. Here, we demonstrate that type I IFNs directly inhibit co-stimulation-dependent Treg proliferation and activation both in vitro and in vivo during acute infection with lymphocytic choriomeningitis virus (LCMV). This inhibition is cell-intrinsic and preferentially targets CD62LloCD44hi effector/memory Treg cells. Moreover, loss of the type I IFN receptor specifically in Treg cells results in functional impairment of virus-specific CD8 + and CD4+ T cells and inefficient viral clearance. Together, these data indicate that inhibition of Tregs by IFNs during acute viral infection is necessary for the development of robust anti-viral T cell responses. Because overexpression of type I interferons is associated with the development of many autoimmune disorders, we also asked how chronic IFN overexpression regulates Treg cell activity and how this contributes to immune dysfunction during IFN-associated autoimmune diseases. Using a well-established in vivo assay of Treg cell function, we show that Treg cell function is impaired in mice that chronically overproduce type I IFNs due to loss of the DNA exonuclease Trex1. Treg cell dysfunction completely depended on type I IFN signaling in T cells. Although IFN overexpression directly inhibited Treg cell proliferation and activation, this direct inhibition was not required for Treg cell dysfunction. Rather, chronic IFN expression directly promoted the expansion of conventional T cells, and inflammatory disease was completely dependent on IFNaR signaling in conventional T cells. Thus, chronic IFN expression drives inflammatory disease by impairing Treg cell function in vivo, both directly and indirectly through its effects on conventional T cells.

Display device color management and visual surveillance of vehicles

Srivastava, Satyam Purdue University 2011 해외박사(DDOD)

Digital imaging has seen an enormous growth in the last decade. Today users have numerous choices in creating, accessing, and viewing digital image/video content. Color management is important to ensure consistent visual experience across imaging systems. This is typically achieved using color profiles. In this thesis we identify the limitations of profile-based color management systems and propose an alternative system based on display device models and look-up tables (LUT). We identify techniques to design LUTs which are optimal in terms of color reproduction accuracy under resource constraints. We show that a LUT-based color management system is more accurate and memory-efficient than a comparable ICC profile-based system. Visual surveillance is often used for security and law enforcement applications. In most cases the video data is either passively recorded for forensic applications or is remotely monitored by human operators. We propose the use of image and video analysis techniques to assist the operators. This would reduce human errors due to fatigue, boredom, and excess information. We describe a video surveillance system to observe vehicular traffic from a standoff range and detect anomalous behavior by analyzing the motion trajectories. We also extract physical information (such as make, tire size, and body type) which can help determine the "normal" behavior. The operator can also use this information to uniquely identify/describe individual vehicles. We describe low complexity techniques to perform the above analyses and show their effectiveness on real traffic videos.

Dynamic friction measurement and modeling at the mirco/nano scale

Srivastava, Abhishek University of California, Santa Barbara 2006 해외박사(DDOD)

An experimental characterization of friction forces between asymmetric surfaces in the micro-regime is performed by exploring the lateral scratching functionality of a traditional nanoindenter (Hysitron Triboindenter) to determine friction properties at low velocities. Classical friction experiments are reproduced using the triboindenter with high repeatability. It is observed that real-time depth measurements closely follow the Hertzian prediction. Friction spikes with magnitude dependent upon the desired velocity input function (higher peak values in reduced rise times) are observed, indicating an onset of kinetic friction effects even before the motion begins. Anisotropy is observed between surfaces of different materials with stick-slip occurring only at specific relative orientations. Experimental estimation of the control parameters of the triboindenter is combined with the LuGre model for describing friction for simulation of the experiments. Data obtained from a tribo-rheometer is least-square fitted to the LuGre model and modifications to the model that describe the friction behavior more accurately and provide a physical basis for it are investigated. Comparison experiments are performed with the tribo-rheometer and the surface force apparatus (SFA) using similar material and scratch parameters to determine the validity of the triboindenter as a nano/micro-scale friction testing tool. Limitations of the triboindenter as a tribological tool and methods to overcome some of them are discussed. Advanced equipment for expanding the current range of the triboindenter which allow experimentation from the nano to the macro scale are tested. Novel methods for scaling friction are described along with possible applications of such tribological investigations in MEMS and nano-scale devices.

Dense phase gas-solid flows in circulating fluidized beds

Srivastava, Anuj Princeton University 2002 해외박사(DDOD)

Circulating fluidized beds (CFBs) are widely used in the chemical industry for fluid catalytic cracking and combustion applications. It is well known that under certain operating conditions, flow instabilities involving the entire CFB system can arise, leading to a breakdown in operation. However, the origin of such instabilities as well as the minimum physics needed in a mathematical model to capture these is not well understood. In this research such CFB loop instabilities have been experimentally and theoretically investigated. The flow behavior of gas-Geldart type A particle mixtures in a pilot scale CFB under stable and unstable operating conditions has been examined. Stable dense phase flow is established at low aeration rates in the standpipe. At high aeration rates the flow becomes unstable, manifesting low frequency oscillations in the flow characteristics. Our results suggest that, under the conditions explored in the present study, this instability originates in the standpipe when the effect of frictional interactions becomes negligible. Any attempt to model this instability should consider the interaction between the various components of the CFB circuit, wall friction and the compressive yield stress of the particle assembly in the standpipe. Careful measurements of pressure drop and bed height during fluidization-defluidization cycles in beds of different diameters have been used to quantitatively determine the compressive yield stress as a function of particle volume fraction for XL particles used in the CFB experiments. Analysis of the standpipe data reveals that the support provided by the standpipe wall can be estimated quantitatively from the standpipe holdup data and the estimated compressive yield stress. A theoretical study of the flow of a gas-Geldart type A particle mixture in a uniformly aerated standpipe system has been presented. The theory is based on a one-dimensional treatment of the volume-averaged equations of motion. It is able to qualitatively capture many of the gas-particle flow features observed in our CFB experiments, such as the effect of aeration on the flow behavior and the operating conditions under which instabilities in the CFB can occur. A frictional model for the rheology of a compressible granular material has been developed and numerical issues associated with its implementation in a CFD framework have been explored. The model is shown to capture the general features of dense phase flows in two test cases in a qualitatively correct manner.