In general, researchers use hand-crafted methods or combine with the deep learning to solve the problem of Pedestrian Detection. In this paper, this problem can be implemented in the purely convolution neural network. Region Proposal Network, proposed by the algorithm for objects detection could be modified and applied on the pedestrian detection. After getting feature maps from the pretrained model, feed them into the new model and train by using Tensorflow as the deep learning framework, we can get the predicted bounding boxes that contain the pedestrians. This method is efficient and can reach the accuracy around 80 percent.

This dissertation presents an investigation on the topic by using surface polaritons to assist electromagnetic wave propagation and radiation, thus used for heat transfer and thermal radiation control. Surface polaritons are collective oscillations of surface charge at metal-dielectric interfaces (surface plasmon polaritons, SPP), and the surface of polar materials (surface phonon polaritons, SPhP). The excited surface polaritons can either propagate along the surface or be confined locally as resonance depending on the interface morphology. We investigated the propagation behavior of SPhP on a polar material SiO2. We designed a SiO2 nanoribbon (NR) waveguide to control the mode size of the surface phonon polaritons and its efficient coupling to thermal reservoirs through numerical simulations. An increased thermal conductivity of as much as 34% over its well-known phonon thermal conductivity limit is demonstrated theoretically and experimentally. This could be an additional channel for heat transfer, complementing the three traditional mechanisms: convection, conduction, and radiation. The localized resonance behavior of surface polaritons can be used for radiation spectrum engineering. We designed a 3D singular metasurface by using the idea of compress dimensions with singularity structures and investigated the optical spectrum numerically and experimentally. By exclusively utilizing only SiO2 in the design of singular metasurfaces, we achieved an experimental demonstration of a longwave infrared (LWIR) emittance of 96.5%. The structure only needs to be prepared with dry etching and wet etching on SiO2 wafer, no complex compounds and structures. It’s cheap, stable, and capable of handling harsh environment which makes it a perfect platform for radiative cooling. In addition to singular metasurface, we also introduce a novel approach as molten metamaterial (MMM) for thermal radiation control at high temperatures. With numerical simulations and optimized structural design, a narrowband emissivity peak at λ=1.6 um with a quality factor of ~5.6 and maximum emissivity >98% were observed, well-aligned with the center wavelength of blackbody radiation at T=1500^o C. This demonstrates the effectiveness of utilizing the MMM to tailor thermal emission spectrum solely by geometries. More experimental demonstrations need to be done, but it already has a good start with a great potential.

In a typical minute of a day in 2018, the Internet generates 3,138 terabytes of traffic, Twitter users send 473,000 tweets, and two million snaps are sent on Snapchat. By 2020, it is estimated that for each person on earth, 1.7 MB of data will be created every second on the average. Due to the large volumes of Big Data, efficient search methods and analytics are required to explore such data. Thus, there is a clear need for Big Data management system, such as Apache AsterixDB, to enable users and applications to search to explore Big Data.

Initiated in 2009, the AsterixDB project integrated ideas from three distinct areas - semi-structured data, parallel databases, and data-intensive computing - to create an open-source software platform that scales on large, shared-nothing commodity computing clusters. AsterixDB currently provides various types of index, such as B+-tree, R-tree, and inverted indexes to fetch data efficiently. Also, as the problem of supporting similarity queries has become increasingly important in many applications, AsterixDB also supports similarity query processing using various metrics and provides an efficient similarity join method. It also provides various search and fundamental analytical functions. It can utilize external third-party libraries using user-defined functions to augment its functionalities. Following the release of the first public open-source version of AsterixDB in 2013, we identified several optics that needed to be explored in depth to enhance AsterixDB further. Those topics are the focus of this thesis.

We first share our memory management experiences in AsterixDB. We describe the original implementation of the system's memory-intensive operations and a set of design flaws (oversights) related to memory management that we found later. We then discuss how we have addressed each of those oversights. We also discuss AsterixDB's memory management at the global level. We believe that future Big Data management system builders can benefit from our memory management experiences. With memory management under control, we next present the design and implementation of index-only query plans in AsterixDB. Use of these plans can boost the performance of an index-based search by several orders of magnitude compared to a scan-based or non-index-only approach. We discuss the challenges that we faced regarding the implementation of index-only query plans in AsterixDB and how we addressed these challenges.

Lastly, considering the importance of similarity query processing for Big Data searches and analytics, we evaluate the performance of similarity query processing in AsterixDB. We compare its approach to several other systems and report the efficacy and performance results.

Investigating the elastic structure at multiple scales of the Earth is crucial for understanding the composition and dynamics of the Earth. Different seismological approaches to extract the information contained in seismic waveforms are presented here. The work consists of three topics: (I) applications of source stacking followed by cross-correlation, (II) modeling seismic anisotropy in the lower mantle, and (III), temporal variation of near-surface seismic velocity and anisotropy in Hokkaido.I. Accurate synthetic seismic wavefields can now be computed in 3D earth models using the spectral element method, which helps improve resolution in full-waveform global tomography. However, computational costs are still a challenge. These costs can be reduced by implementing a source stacking method (Capdeville et al. 2003), in which multiple earthquake sources are simultaneously triggered in only one teleseismic SEM simulation. One drawback of this approach is the perceived loss of resolution at depth, in particular because high-amplitude fundamental mode surface waves dominate the summed waveforms, without the possibility of windowing and weighting as in conventional waveform tomography. This can be addressed by redefining the cost function and computing the cross-correlation wavefield between pairs of stations before each inversion iteration. While the Green’s function between the two stations is not reconstructed as well as in the case of ambient noise tomography, where sources are distributed more uniformly around the globe, this is not a drawback, since the same processing is applied to the 3D synthetics and to the data, and the source parameters are known to a good approximation. By doing so, time windows with large energy arrivals corresponding to fundamental mode surface waves can be separated. Also, applying a weighting scheme can bring out the contribution of overtones and body waves and also can balance the contributions of frequently sampled paths. The approach is computationally very efficient and can help address such questions as model resolution in the presence of noise, and trade-offs between different physical parameters (anisotropy, attenuation, crustal structure etc..) that would be computationally very costly to address adequately when using conventional full waveform tomography based on single-event wavefield computations.

II. By assuming a geodynamical scenario of slab subduction and considering the seismic anisotropy in D” is dominated by crystal preferred orientation, the textural evolution and re- resultant elastic properties can be computed, which may help to explain the seismic anisotropy observations (e.g., McNamara et al. 2002, 2003; Wenk et al. 2011). In previous work, Cottaar et al. (2014) calculated the seismic anisotropy produced in a single mineral system, comparing bridgmanite (MgSiO3 perovskite, Pv) to magnesium post-perovskite (pPv), based on elastic properties and slip systems determined from laboratory experiments and theoretical ab-initio computations. More realistic situations are considered in this study: (a) the polycrystal plasticity model (Lebensohn and Tom ́e 1993) which consists Pv/pPv, cubic calcium perovskite (CaSiO3, CaPv), and cubic periclase (MgO), (b) the forward and reverse Pv-pPv phase transitions during the slabs’ subduction and the subsequent upwelling, (c) partial melting in the deepest portions of the slab at the base of upwelling. To validate and compare the results with seismological observations, the spatial distribution of radial anisotropy described by the parameter ξ = (VSH/VSV)2, and shear wave splitting (SWS) directions and strengths are extracted. III. Here an earthquake-based method (Chen et al. 2017) is applied through interferometry of earthquake coda waves to investigate the temporal changes of elastic properties by using borehole-surface station pairs. The technique provides comprehensive observations not only of S-wave velocity but, additionally, of P-wave and S-wave azimuthal anisotropy between two sensors. The 15-year temporal variations of elastic properties are monitored at the KiK- net station IBUH03 in southern Hokkaido, which experienced the 2003 Mw 8.3 Tokachi-Oki (PGA= ∼ 350cm/s2) and the 2018 Mw 6.7 Hokkaido Eastern Iburi (PGA= ∼ 500cm/s2) earthquakes. The preliminary results show sudden reductions of VSiso and VP after the two major earthquakes considered, and these were subsequently recovered with short-term and long-term recovery rates. Changes in the fast S-wave polarization direction and strength of anisotropy were also detected during the coseismic periods. Furthermore, a clear seasonal trend of elastic parameters can be observed. The results are compared with temporal variations in equivalent water thickness, precipitation, and temperature. A Hudson-Crampin anisotropic effective model is implemented to understand the seasonal crack behavior. The modeling result indicates that crack density and the aspect ratio could play an important role in our observations.

Cancer remains one of the most difficult diseases to treat. In the past half century, cancer treatment has heavily relied on small molecule drugs, which are very potent on killing cancer cells, but lacks the tumor specificity which brings high toxicity against normal cells. For this reason, the improvement and development of new therapeutics that target cancer cells specifically is critical so that toxicity can be minimized for normal cells.

One potential targeting strategy is to exploit the alterations in amino acid synthesis, or salvage pathways displayed by cancer cells. The goal is to target tumors that are sensitive to one of the conditionally essential or the nonessential amino acids. This allows the normal cells to remain unaffected because they have the ability to synthesize sufficient amounts of necessary amino acids to survive. Unfortunately, most enzymes with the potential for amino acid depletion therapy are derived from nonhuman sources, which makes them highly immunogenic. Therefore, for these enzymes to have clinical efficacy, they must be formulated to be delivered in a way that can avoid or delay the immune response. Arginine deiminase (ADI) is a great example of a nonhuman enzyme that depletes a semi-essential amino acid, arginine, in the human body. ADI has shown some potential in treating hepatocellular carcinoma, melanoma, and some mesotheliomas cancer patients.

The scientific significance of this dissertation is to engineer the recombinant native ADI, PEGylated ADI, and nanoparticle platform to encapsulate ADI so that it can be a better drug candidate compared to its native and PEGylated ADI by assessing their pharmacodynamics and pharmacokinetics properties. PEGylating ADI or other enzymes is the current standard formulation that delivers a nonhuman enzyme by creating a shielding layer around the molecule to delay its immune response. If the proposed silica coated liposome encapsulated ADI nanoparticle platform can be synthesized to avoid the immune response completely, it will out-compete the PEGylation strategy. Finally, the comparison between the different formulations will be tested in vitro and in vivo settings to prove this concept.

Lidars are becoming common components for remote sensing in many emerging applications such as autonomous vehicle and facial recognition. To accurately visualize the point cloud for further data processing, metrics such as precision, sampling rate, and linearity are important. Phase detection has been widely used for high precision, however, resolving the ranging ambiguity beyond one carrier period lowers the system acquisition rate. Pulsed direct time-of-flight (ToF) detection, on the other hand, provides high sampling rate with single-shot measurements, but the precision is limited by the walk error. In this work, we present a pulsed-coherent detection to combine the advantages from both the pulsed detection with high sampling rate and the phase detection with high precision. The CW laser source is amplitude modulated by a high RF carrier and a low frequency pulsed envelope modulation. At the receiving end, the segmented time-of-flight detection algorithm measures the phase shift of the RF carrier as fine ToF and counts the arrival time of the low-frequency mask's envelope as intermediate and coarse ToF. This pulsed-coherent lidar can simplify the optical setup while achieving high precision and high sampling rate. Three different receiver prototypes are demonstrated with two test chips: (1) analog-based receiver which has separate paths for coarse and fine detection. The coarse ToF is recorded by detecting the pulse’s edge whereas the fine measurement is measured by phase detection. The proposed post-edge detection with automatic gain control loop suppresses the walk error; (2) DSP-based homodyne receiver which we reuse the same chip of analog-based receiver and collect the data by sub-sampling ADCs. Both the coarse detection and fine detection are calculated in the digital signal processor (DSP). This receiver improves the immunity to PVT variation and inherently aligns the segmented measurements by using the sampling clock as the reference hence achieving better precision at higher sampling rates; (3) DSP-based heterodyne receiver which we take a lower frequency clock as the reference input and use a local phase-locked loop (PLL) to generate the clock for the down-conversion mixer. This LO generation reduces the coupling between the clock and the incoming signal. Also, the two-step down-conversion with digital down-mixing improves the power efficiency and immunity to PVT variation without additional calibration. In addition to the innovation of the architecture, this work also presents some novelties of the circuit implementations: (1) we introduce a narrowband input-matching to improve the noise performance which uses direct wire-bonding between the photodiode bare die to the low noise amplifier; (2) we implement phase-invariant amplifiers to operate the lidar system in wide dynamic range; (3) we build a low phase noise LO generator based on a ring-VCO-based phase-locked loop. Two test chips are fabricated in TSMC 28nm CMOS technology. We successfully demonstrate the lidar system with both 1-D scanning and 2-D scanning. The system achieves <10�m precision with 5-MHz integration bandwidth and 30-�m INL at 2.5-m distance.