Skip to main content
eScholarship
Open Access Publications from the University of California

UC Santa Cruz

UC Santa Cruz Electronic Theses and Dissertations bannerUC Santa Cruz

UC Santa Cruz Electronic Theses and Dissertations

Cover page of Classification of the Sex of Drosophila Suzukii with Pre-Trained Networks

Classification of the Sex of Drosophila Suzukii with Pre-Trained Networks

(2023)

There has been a recent trend to applying deep learning methods compared to shallowmethods for automatic identification of insects. Classification strategies built around al- gorithms with deep learning architectures at their center like YOLO and others require large amounts of data to making learning successful and are often augmented with tens of thousands of images or more to achieve excellent performance. Recent pre-trained models of deep neural networks have significantly reduced the amount of data required to create accurate classification algorithms by ingesting and training on a huge data set different than the target task and using the resulting encoding to transfer information to a new task. This work shows that recent performance gains from models pre-trained on huge data sets are effective as image encoders for the classification of the sex of spotted wing drosophila (SWD). A data set of 676 SWD microscope images is created to evaluate classification models for use in automation of the sterile insect technique (SIT), which requires large amounts of male SWD to be identified and separated. Bi- nary classification models trained on top of image encoding from new models based off of visual transformers [3] pre-trained on over 400 million images with CLIP [2] are able to achieve accuracy as high as 96.7% when trained with LogReg and similar classifiers on augmented data from the SWD image set. Other models pre-trained on the ImageNet data set of 14 million images also performed well, approaching 92% with VGG models and 90% with MobileNetV2 model. Image segmentation of the data set is then inves- tigated as a source of corroboration for the identification of the morphological features responsible for classification, and an out-of-distribution data set is collected to evaluate classification and segmentation results on more diverse and difficult examples. While robust identification of features special to SWD remains, classification accuracy is not a guarantee on data which differs substantially from the factory or laboratory setting on which it is trained and additional data may be needed for training on use-cases outside of SIT such as for applications on the farm or for automated identification in insect traps. This emphasizes a fact which is not elaborated on for many insect detection models in the literature: that their models are not likely robust in situations where the data is significantly OOD and for situations which may not be adequately covered with- out specialized augmentation methods or additional data. Nonetheless results indicate that pre-trained models have advanced to the point where they can play a central role in securing the food supply from potentially billions of dollars of damages every year from pests such as SWD.

Cover page of Detection methods for discovering Evaporating Primordial Black Holes in modern Gamma-ray Telescopes

Detection methods for discovering Evaporating Primordial Black Holes in modern Gamma-ray Telescopes

(2023)

The potential of directly observing a primordial black hole (PBH) explosion carries immense implications for our understanding of the universe, from cosmology to particle physics. While the existence of PBHs as a candidate for dark matter has been theorized for decades, direct detection of an evaporating PBH would provide invaluable insights into yet-undiscovered high-energy particles and dark radiation. In this thesis, we review the role of PBH as a dark matter candidate, including formation mechanisms and an overview of the state of the field. We model the detection limits for PBH sources in the gamma photon spectrum for a number of modern telescopes and discuss constraints due to velocity dispersion, multi-spectral characteristics and lightcurve evolution. We investigate also the possibility and effect of dark particle radiation arising from dark degrees of freedom in black hole evaporation, and the possibility of a multi-spectral afterglow from PBH evaporation products. Lastly, we apply these novel constraints to the Fermi mission catalogs and produce several candidates at a variety of different possible evaporation stages and distances.

Cover page of Essays on Belief-Updating and Decision-Making in Financial Markets

Essays on Belief-Updating and Decision-Making in Financial Markets

(2023)

This dissertation contains three essays broadly related to financial markets, with an emphasis on decision-makers’ belief-updating and decision-making.

Chapter 1 studies subjects’ belief-updating when they face an uncertain event accompanied by two independent signals in the laboratory. The “Average model” is introduced and compared with other important models for the goodness off fit. At theindividual level, the results are mixed. Some subjects behave close to one of the model’s predictions, while some behave close to another model’s predictions. The average model outperforms the Bayesian model at the aggregate level in predicting subjects’ posterior beliefs. No clear evidence indicates that subjects’ posterior beliefs converge to the Average or Bayesian model’s predictions over time.

Chapter 2 studies the impact of motivated beliefs, the phenomenon that people believe what they want to believe, on market performance in a laboratory market for a state-contingent common value asset. Motivated beliefs are induced so that traders have polarized preferences over the states. The main findings are that (i) these induced motivated beliefs do not have a significant impact on overall market efficiency, but (ii) they do impact traders’ final asset holdings and belief updating processes, and (iii) the induced polarization persists after receiving private signals and trading in the market. Other findings suggest that a more intense financial stake might improve market efficiency. The induced ego-relevant motivation is significantly stronger than the homegrown motivation to believe in higher payoffs.

In Chapter 3, joint work with Daniel Friedman and Thomas Bowen, we study the impact of traders’ overconfidence on market performance. How does trader overconfidence (judgemental or self-enhancement) affect their performance in asset markets, and overall market quality? Conversely, how does market participation affect traders’ overconfidence? To address such questions, we build a laboratory asset market in which human participants receive private information of varying precision and then trade an asset that pays a single state-contingent dividend. Among other results, we find that greater trader overconfidence can improve price efficiency in some environments, but not in the most realistic environment with experienced traders and ambiguous mixed information precision. In that environment, overconfidence reduces trader profits. We detect no substantial impact of market exposure on trader overconfidence.

Cover page of “See The Natives”: Indigenous Visual Culture at the 1894 California Midwinter International Exposition

“See The Natives”: Indigenous Visual Culture at the 1894 California Midwinter International Exposition

(2023)

ABSTRACT“SEE THE NATIVES”: INDIGENOUS VISUAL CULTURE AT THE 1894 CALIFORNIA MIDWINTER INTERNATIONAL EXPOSITION Christina Hellmich Though the 1894 California Midwinter International Exposition (or CMIE) in San Francisco is often considered a minor regional exposition and is typically omitted from scholarly consideration of world’s fairs, it was significant for raising the profile of the city as Western expansion came to a close, serving to promote the colonialist interests of settlers in the West. This study investigates the exhibition of Indigenous culture at CMIE, as configured through villages complete with “Natives” (Indigenous people from inside and outside US borders), putatively depicting their ways of life. The villages shaped and reinforced understandings of race and national identity and made a case for regional and global imperialism. Through the legacy of the Expo’s photographic and textual archive, we can explore some of the strategies and counterstrategies deployed at the CMIE and their resulting meanings. Like theatrical sets, the villages were conceived by their organizers as performance spaces to represent cultures—merging the authentic, the adapted, and the fabricated—to create an encounter for visitors with the desired message and financial remuneration. Examples drawn from the Native Hawaiian, Pacific Islander, Fon, and Native American villages illustrate how the imaging and spectacle of Indigenous participants were powerful tools demarcating difference and enabling the United States and European nations to define their national identities against the racial and cultural stereotypes that they created of Indigenous people. These, in turn, played a larger role in national debates about US expansion and economic imperialism. While it can be shown that performances at the CMIE were staged in hopes of cementing ideas about settler colonialism and White supremacy put forward by village managers and organizers, the displays were nonetheless subject to unanticipated mutability and expansion of viewpoints promulgated by the Indigenous participants. Through their biographies, it is possible to move beyond the generalizations and stereotypes applied to Midwinter Fair participants to reveal how they were responding to financial and political instability in their homelands and larger cultural debates to assert power over their bodies, identities, and representation at the fair.

Cover page of openCArM: open source camera array microscopy for Biology

openCArM: open source camera array microscopy for Biology

(2023)

The need for experimental replicates means biology research labs are oftengrowing many cell cultures in parallel. Monitoring the health of these cultures as they grow is important for avoiding costly interruptions in the research pipeline. Common approaches involve removing culture plates from incubator environments to view them under a microscope. This manipulation can stress cells and end up being disruptive to healthy growth environments. To facilitate non-disruptive automated cell culture monitoring, an open source framework for the design and deployment of parallel multi-channel microscopes was developed. Open Camera Array Microscopy (or openCArM) supports the design of multi- well parallel imaging systems composed of independent autonomous camera units. The framework defines module types that can be assembled together and config- ured for a given use case. One such configuration (the "Picroscope") performs in-incubator, longitudinal imaging studies on 24 well cell culture plates. "Picro- scope" devices have been used in experiments with samples ranging from Xenopus Tropicalis frog embryos to human cortex brain organoids. openCArM systems have a modular design that allows them to be built around other cell biology automation devices. Experiments have been run using the "Picroscope" combined with the "Autoculture" system, an automated cell culture feeding platform. To support these devices, a connected lab equipment management system was developed to provide experiment tracking, data analysis and equipment control. openCArM devices can be made out of low cost 3D printed materials and off-the-shelf imaging hardware, representing a significant step forward for the accessibility of cell culture automation systems.

Cover page of Evidence-based gamification and mnemonics for logographic writing systems.

Evidence-based gamification and mnemonics for logographic writing systems.

(2023)

People have many reasons to want to learn a foreign language: some want to enjoy foreign media in its original form, some need it for business or travel, personal enrichment, and so on. Roughly 1/6 of Earth's population's learn languages with logographic writing systems (LWS) such as Chinese, Japanese and Korean as their native tongues. These peoples harbor an enormous cultural, economic and academic wealth. As a result, these languages are a popular choice for a variety of learners ranging from serious businessmen to K-pop and anime fans. Whatever the motivation, learning additional languages carries a slew of cognitive benefits and should be encouraged.

Maintaining the motivation and persistence needed to learn a language is a difficult task, and the problem is amplified for learners with alphabetic roots tackling an LWS language. One needs to memorize around 2000 characters to be considered literate. This is routinely done through rote memorization, and the prospect discourages many would-be learners. When it comes to LWS as a foreign language, there is a divide between conversational fluency and literacy unlike that in any alphabetic language. My research aims to help bridge this gap through development and evaluation of LWS learning games. In this, I make a point to rely on insights from LWS teaching researchers, best practices of educational game design, and advice from the videogame industry experts.

My approach brings innovation in two areas:

I contribute to LWS instruction by introducing two LWS-learning game designs that focus on production tasks, while the vast majority of games out today target recall only.

I contribute to educational game design field by exploring the use of music as a mnemonic for learning, and conceptualizing the potential future use of notorious commercial game engagement and retention mechanics in an educational context.

Cover page of Search for Higgs Bosons Produced via Vector Boson Fusion and Decaying to a Pair of $b$-quarks in Association with a High-Energy Photon in the ATLAS Detector at the Large Hadron Collider

Search for Higgs Bosons Produced via Vector Boson Fusion and Decaying to a Pair of $b$-quarks in Association with a High-Energy Photon in the ATLAS Detector at the Large Hadron Collider

(2023)

A search for the Standard Model Higgs boson produced in association with a high-energy photon is performed using 132 fb$^{-1}$ of $pp$ collision data at $\sqrt{s}={13}$ TeV collected with the ATLAS detector at the Large Hadron Collider. This dissertation presents a complete analysis of the vector boson fusion production mode of the Higgs boson, which is a particularly powerful channel for studying the $H(\rightarrow b\bar{b})+ \gamma$ final state because the photon requirement greatly reduces the multijet background and because the Higgs boson decays primarily to bottom quark-antiquark pairs. Extending and updating background Monte Carlo samples, training a neural network to distinguish between signal and background events, and optimizing binned-liklihood signal and background model fitting techniques are new strategies used in this analysis.

Cover page of Building Practical Statistical Relational Learning Systems

Building Practical Statistical Relational Learning Systems

(2023)

In our increasingly connected world, data comes from many different sources, in many different forms, and is noisy, complex, and structured. To confront modern data, we need to embrace the structure inherent in the data and in the predictions. Statistical relational learning (SRL) is a subfield of machine learning that provides an effective means of approaching this problem of structured prediction. SRL frameworks use weighted logical and arithmetic expressions to easily create probabilistic graphical models (PGMs) to jointly reason over interdependent data. However, despite being well suited for modern, interconnected data, SRL faces several challenges that keep it from becoming practical and widely used in the machine learning community. In this dissertation, I address four pillars of practicality for SRL systems: scalability, expressivity, model adaptability, and usability. My work in this dissertation uses and extends Probabilistic Soft Logic (PSL), a state-of-the-art open-source SRL framework.

Scalability in SRL systems is essential for using large datasets and complex models. Because of the complex nature of interconnected data, models can easily outgrow available memory spaces. To address scalability for SRL, I developed methods that more efficiently and intelligently instantiate PGMs from templates and data. I also developed fixed-memory inference methods that can perform inference on very large models without requiring a proportional amount of memory.

Expressivity allows SRL systems to represent many different problems and data patterns. Because SRL uses logical and arithmetic expressions to represent structured dependencies, SRL frameworks need to be able to express more than just what is represented by feature vectors. To address expressivity for SRL, I created a system to incorporate neural models with structured SRL inference, and expanded the interpretation of PSL weight hyperparameters to include additional types of distributions.

Model adaptability is the ability of SRL frameworks to handle models that change. A changing model can be as simple as a model that has its hyperparameters updated, or as complex as a model that changes its structure over time. To address model adaptability for SRL, I developed new weight learning approaches for PSL, and created a system for generalized online inference in PSL.

Usability make SRL frameworks easy for people to use. Because of the need to model structural dependencies, SRL frameworks are often harder to use when compared to more common machine learning libraries. To address usability for SRL, I have created a new SRL framework that removes the tight coupling between the different components of the SRL pipeline that is seen in other SRL frameworks and allows the the recreation of exiting SRL frameworks along with the creation of new SRL frameworks using the same common runtime. Additionally, I developed a visual model inspector for analyzing and debugging PSL models.

Cover page of An Open Source Real-time Controller for Resource-constrained Autonomous Vehicles and Systems

An Open Source Real-time Controller for Resource-constrained Autonomous Vehicles and Systems

(2023)

The use of autonomous systems is burgeoning in the world for applications in many fields from scientific, industrial, to military. At the same time, advances in semiconductor technology have enabled ever smaller, complex, and use-specific microprocessors and microcontrollers. This work details the design and implementation of an open source real-time hardware controller for resource-constrained autonomous vehicles and systems. It is intended to be integrated inside a distributed control architecture consisting of the real-time hardware controller, a guidance and navigation computer, and an edge tensor processing unit for machine learning inferences. While the latter two processors are commercially available, a dedicated, modular real-time controller is not, providing the motivation for this work. To demonstrate the versatility of our open source real-time controller we present several use cases including a ground vehicle, marine vessel, quadcopter, and fixed-wing aircraft. The power of the distributed architecture is the ability to solve complex sensing, guidance, navigation, and control challenges even in resource-constrained systems. One such challenge is the simultaneous localization of an autonomous system while mapping an environment. In this work we develop the components of a novel hybrid sensor combining a visual camera and LiDAR sensor that is mounted on the ground vehicle. This sensor is trained to recognize landmarks in the environment using object detection frameworks and deployed on the edge tensor processing unit. At the same time, the LiDAR sensor provides range and bearing information for objects within its field of view. By combining the two we can get fast detections of arbitrary landmarks in the environment as well as determine their position relative to the sensor, thus enabling simultaneous localization and mapping functionality.

  • 1 supplemental PDF
Cover page of Still Searching: research and extension in California organic no till vegetable production

Still Searching: research and extension in California organic no till vegetable production

(2023)

As climate change presents more and varied challenges for food production, there is a need for novel systems that can balance social and ecological outcomes. No till practices have shown incredible promise providing important ecological benefits, but these systems are still heavily reliant on chemical herbicides and fertilizers and have been limited in adoption to major commodity crops in humid regions without water limitations. Organic agriculture has proven to be a viable production system across a wide range of crops while excluding synthetic pesticides and chemicals that create downstream effects for human and non-human communities. There has been little success in developing the “holy grail” of organic no-till farming, especially for nutritionally important vegetable crop systems in California. I explore the history of organic no till research (Chapter 1), highlighting specific challenges and opportunities from regional syntheses of organic no till production in commodity crops as well as the slowly growing body of literature on organic no-till vegetable production. I then review the results of a 3-year field trial exploring a novel low-reside organic no till production system on yield and nutrient dynamics (Chapter 2). Finally, I explore the cooperative extension service (Chapter 3) and discuss why it has failed to support ecological innovations given its contested history. This work argues that while research on organic no till systems is still at a nascent stage, there are a number of meaningful research pathways to pursue: 1) a focus on basic agronomic challenges, 2) suitability of cover crops and specific cash crop/cover crop combinations, and 3) long-term studies without cash crops to assesses soil physical and chemical properties. Further, given the nature of highly commodified California vegetables, cooperative extension, despite its inconsistent track record, has an important role in supporting newer models of social learning, adaptive research and university/grower partnerships that are needed to support this and future endeavors.