We present recent work on a prototype compact neutron generator for associated particle imaging (API). API uses alpha particles that are produced simultaneously with neutrons in the deuterium-tritium (2D(3T,n)4 alpha) fusion reaction to determine the direction of the neutrons upon exiting the reaction. This method determines the spatial position of each neutron interaction and requires the neutrons to be generated from a small spot in order to achieve high spatial resolution. The ion source for API is designed to produce a focused ion beam with a beam spot diameter of 1-mm or less on the target. We use an axial type neutron generator with a predicted neutron yield of 108 n/s for a 50 muA D/T ion beam current accelerated to 80 kV. The generator utilizes a RF planar spiral antenna at 13.56 MHz to create a highly efficient inductively-coupled plasma at the ion source. Experimental results show that beams with an atomic ion fraction of over 80percent can be obtained while utilizing only 100 watts of RF power in the ion source. A single acceleration gap with a secondary electron suppression electrode is used in the tube. Experimental results, such as the current density, atomic ion fraction, electron temperature, and electron density, from ion source testing will be discussed.

## Type of Work

Article (38) Book (0) Theses (35) Multimedia (0)

## Peer Review

Peer-reviewed only (58)

## Supplemental Material

Video (0) Audio (1) Images (0) Zip (1) Other files (2)

## Publication Year

## Campus

UC Berkeley (1) UC Davis (1) UC Irvine (0) UCLA (61) UC Merced (0) UC Riverside (0) UC San Diego (11) UCSF (4) UC Santa Barbara (1) UC Santa Cruz (0) UC Office of the President (1) Lawrence Berkeley National Laboratory (5) UC Agriculture & Natural Resources (0)

## Department

Department of Statistics, UCLA (12) School of Medicine (9) Research Grants Program Office (RGPO) (1)

## Journal

## Discipline

## Reuse License

BY - Attribution required (1)

## Scholarly Works (73 results)

To achieve the required damping time in the main damping rings for the Next Linear Collider (NLC), a wiggler will be required in each ring with integrated squared field strength up to 110 T^2m. There are concerns that nonlinear components of the wiggler field will damage the dynamic aperture of the ring, leading to poor injection efficiency. Severe effects from an insertion device have been observed and corrected in SPEAR 2. In this paper, we describe a model that we have developed to study the effects of the damping wiggler, compare the predictions of the model with actual experience in the case of the SPEAR 2 wiggler, and consider the predicted effects of current damping wiggler design on the NLC main damping rings.

The CRISPR base editors are programmable DNA editing systems that induce single-nucleotide changes in the DNA using a fusion protein containing a catalytically defective Cas9, a cytidine or adenine deaminase, and an inhibitor of base excision repair. This genome editing approach has the advantage that it does not require the generation of double-stranded DNA breaks or a donor DNA template. The single guide RNAs (sgRNAs) enables the precise editing at the designed region using CRISPR base editor. Different sgRNAs have largely different efficacy for base editing, and computational prediction can facilitate the optimized design of sgRNAs with high editing efficacy, sensitivity and specificity. Here we present a convolutional neural network-based approach to predict the sgRNAs editing efficiency for CRISPR base editing. Firstly, we designed a large-scale sgRNA library of over 7,000 sgRNAs to introduce pre-mature stop codons into essential genes in yeast, where the yeast would drop out from the population if the sgRNA works efficiently due to the disruption of essential genes. The base editing efficiency of the 7,000 sgRNAs was measured by the log ratio change of sgRNA abundance at 72 h after the editing induction and at the beginning. We built a CNN model using the sgRNA sequence and the surrounding DNA context as the training input to predict the editing efficacy of any given sgRNA sequences. With architecture and parameter tuning, the CNN model surpassed the machine learning approaches tested. In addition, the CNN model fully automated the identification of sequence that may affect sgRNA editing efficacy in a data-driven manner.

The use of responses from questionnaires is ubiquitous in social and behavioral science research. One side effect of using such data is that researchers must often account for item level missingness. Multiple imputation is one of the most widely used missing data handling techniques, wherein missing data are replaced by plausible values from the their proper posterior distribution given the observed data. Instead of the standard procedure in structural equation modeling (SEM), which requires researchers to fit their model to imputed data sets as many times as the number of imputations and then combine parameter estimates and standard errors at the end, we propose a new and simpler approach that is computationally more convenient. It has a number of additional benefits such as the availability of fit indices. Motivated by Lee and Cai (2012), who proposed an alternative method for statistical inference under MI in SEM with continuous variables, we extend their approach to the case of categorical variables.

Within the context of ordered categorical data, the main idea is summarized as follows. Assume we have thresholds and polychoric correlations computed from M imputed data set. Our goal is to perform estimation and inference with these M different thresholds and polychoric correlations. We can easily average the thresholds and polychoric correlations; however, the weight matrix for obtaining the correct statistic in CSEM requires reflecting the between-imputation variance on top of simple averaging of asymptotic covariance matrices of the thresholds and polychoric correlations. Finally, applying Browne (1984)’s Proposition 4 leads us to obtain the correct test statistic, ˜TB. We further consider ˜TYB, a small-sample adjustment of ˜TB (Yuan & Bentler, 1997). We demonstrate our proposed statistics performance and their power to detect model misspecification via simulation studies. In addition, we illustrate our findings with two empirical data sets.

Ion source development plays an important role for improving and advancing the neutron generator technology used for active interrogation techniques employed by the Department of Homeland Security. Active neutron interrogation using compact neutron generators has been around since the late 1950's for use in oil well logging. However, since the September 11th, 2001 terrorists attack, much attention has been paid to the field of active neutron interrogation for detecting hidden explosives and special nuclear materials (SNM) in cargo and luggage containers through the development of effective and efficient radioactive sources and detectors. In particular, the Associated Particle Imaging (API) method for detecting and imaging explosives is of great interest New compact neutron generators will help to enhance the capabilities of existing threat detection systems and promote the development of cutting-edge detection technologies.

The work performed in this thesis includes the testing of various ion source configurations and the development and characterization of an inductively coupled radio frequency (RF) ion source for use in compact neutron generators. These ion source designs have been investigated for the purpose of D-T neutron generation for explosive detection via the Associated Particle Imaging (API) technique. API makes use of the 3.5 MeV alpha particles that are produced simultaneously with the 14 MeV neutrons in the deuterium-tritium (^{2}D(^{3}T,n)^{4}α) fusion reaction to determine the direction of the neutrons and to reduce background noise. The Associated Particle Imaging neutron generator required a beam spot of 1-mm or less in diameter at the target in order to achieve the necessary spatial image resolution. For portable neutron generators used in API, the ion source and target cannot be water-cooled and the power deposited on the target must be low. By increasing the atomic ion fraction, the ion beam can be used more efficiently to generate neutrons, resulting in a lower beam power requirement and an increased lifetime of the alpha detector inside the acceleration column. Various source configurations, antenna design, and permanent magnet placement have been investigated so as to develop an ion source which could provide high monatomic deuterium species and high current density at relatively low RF powers (less than 200 W).

In this work, an RF ion source was developed that uses an external, planar, spiral antenna at 13.56 MHz with a quartz source body and side multi-cusp magnets to generate hydrogen isotope plasmas with high mono-atomic ion species (> 80%) while consuming only 150 watts of power and operating under 10 mTorr of gas pressure. A single acceleration gap with a secondary electron suppression electrode are used in the tube. Experimental measurements of the ion source plasma parameters including ion current density, atomic ion fraction, ignition and operating pressures, electron temperature, and electron density are presented along with a discussion on the ion optics and engineering challenges. It is shown that the measured neutron yield for the developed D-D neutron generator was 2 × 10^{5} n/s, which scales to 8 × 10^{7} n/s for D-T operation. In addition, initial measurements of the neutron generator performance including the beam spot size, associated particle detection, and neutron tube (without pumping) operation will be discussed. Some suggestions for future improvement are also presented in this dissertation.

Developing statistical models and associated learning algorithms for the rich visual patterns in natural images is of fundamental importance for computer vision. More importantly, the endeavor has the potential to enrich our treasured collections of statistical models and expand the already vast reach of machine learning methodologies. Generative models enable us to learn useful features and representations from the natural images in an unsupervised manner. The learned features and representations can be more interpretable and explicit than those learned by the discriminative models, especially if the learned models are sparse. The objective of this dissertation is to learn probabilistic generative models for representing visual patterns in natural images.

In this dissertation, we first develop a sparse FRAME model as a generative model for representing natural image patterns. The model is an inhomogeneous and sparsified version of the original FRAME (Filters, Random field, And Maximum Entropy) model. More specifically, it is a probability distribution defined on the image space which combines wavelet sparse coding and Markov random field modeling. We propose two different algorithms to learn this model. The first is a two-stage algorithm that initially selects the wavelets by a shared sparse coding algorithm and then estimates the weight parameters by maximum likelihood via stochastic gradient ascent. The second approach utilizes a single-stage algorithm that uses a generative boosting method combined with a Gibbs sampler on the reconstruction coefficients of the selected wavelets. Our experimental results show that the proposed sparse FRAME model can not only learn to generate realistic images of a wide variety of image patterns, but can also be used for object detection, clustering, codebook learning, bag-of-word image classification, and domain adaptation.

We further propose a hierarchical version of FRAME models that we call generative ConvNet. The probability distribution of the generative ConvNet model is in the form of exponential tilting of a reference distribution, and the exponential tilting is defined by ConvNet that involves multiple layers of liner filtering and non-linear transformation. Assuming re-lu non-linearity and Gaussian white noise reference distribution, we show that the generative ConvNet model contains a representational structure with multiple layers of binary activation variables. The model is piecewise Gaussian, where each piece is determined by an instantiation of the binary activation variables that reconstruct the mean of the Gaussian piece. The Langevin dynamics for synthesis is driven by the reconstruction error, and the corresponding gradient descent dynamics converges to a local energy minimum that is auto-encoding. As for learning, we show that the contrastive divergence learning tends to reconstruct the observed images. We also generalize the spatial generative ConvNet to model dynamic textures by adding the temporal dimension. The spatial-temporal generative ConvNet consists of multiple layers of spatial-temporal filters to capture the spatial-temporal patterns in the dynamic textures. Finally, we show that the maximum likelihood learning algorithm can generate not only vivid natural images but also realistic dynamic textures.

We lastly investigate a connection of the proposed models to auto-encoders. We show that the local modes of both the sparse FRAME model and the generative ConvNet are represented by auto-encoders, with explicit encoding of the data in terms of filtering operations, and explicit decoding that generates the data in terms of the basis functions that corresponds to the filters. We call these auto-encoders the Hopfield auto-encoders because they describe the local energy minima of the models. We develop learning algorithms to learn those models by fitting Hopfield auto-encoders. We show that it is possible to select wavelets and estimate weight parameters for sparse FRAME models by fitting Hopfield auto-encoders. Moreover, meaningful dictionaries of filters can be obtained by learning hierarchical Hopfield auto-encoders for generative ConvNet. Without MCMC, the Hopfield auto-encoder has the potential to tremendously accelerate learning that is crucial for big data.

Linear mixed effects models have been widely used in different disciplines and have become a large research field of Statistics. With the development of science and technology, a large amount of variables are always available to choose for a model and it is necessary to control the numbers of variables to avoid the overfitting problem and use the most efficient way to explain data. Most methods published pay more attention to the selection and estimation of fixed effects but it is meaningful to get a deep insight into variable selection for random effects. Some adjustments have been made in this thesis to obtain the specific methods for variable selection on random effects model based on reviews of some classic or latest methods for variable selection on mixed effects model. These methods and algorithms have been applied on some simulation data and compared through changes on number of subjects and observations. Additionally, these methods have been applied into a real world dataset to study how some effects will influence the democracy index among different countries.

Data centers use large numbers of hard drives as data storage devices and it is an increasing challenge to maintain the reliability of the storage system as the number of the hard drives increases exponentially. Manual monitoring does not seem to be efficient for large scale storage systems. Typically, the distributions of healthy hard drives and failed hard drives are highly imbalance. In addition, the size of the training data is large for large scale storage systems. The existence of such challenges makes the hard drive failure prediction problem interesting. In this thesis, several classification models are applied to the hard drive S.M.A.R.T. data from 34,970 hard drives for failure prediction, and the results are compared. Based on the analysis, XGBoost provides the best overall prediction result and it is able to process the data efficiently.

This thesis is to meet the needs of developing automation process on defective testing in the

hard disk drives production process, focusing on machine learning and articial intelligence.

The objective is to to predict the defectives and improve the accuracy rate by using the tree

based algorithm and neural networks. The powerful models can help manufacturing process

improving time and labor efficiency.