Search

The emerging demands for healthcare where access is limited due to political, environmental, or socio-economic factors have been driving research into bio-medical devices that perform in both diagnostic and therapeutic roles at lower costs and greater accessibility. Paper microfluidic devices are used in many applications, particularly medical diagnostics and offer an excellent combination of utility and low cost making them particularly valuable in resource-limited applications and point-of-care usage across a wide variety environmental conditions. Microfluidic biological diagnostics continue to mature as researchers discover new ways to exploit the technological possibilities, and address liabilities. The increasing complexity of paper-based microfluidic devices beyond home pregnancy tests is driving the need to produce new tools and methodologies that enable more robust biological diagnostics and potential therapeutic applications. However, the process of developing new paper microfluidic devices is limited due to having to manually design and fabricate designs to research. Often, researchers must design scores of different devices to find a combination of parameters that functions as expected. In this work, a novel software framework to support automated development of paper-based microfluidic devices is introduced to facilitate both research and fabrication to accelerate the investigative process and reduce material utilization and manpower. Unlike to existing lab-on-a-chip technologies, paper-based microfluidics differs in terms of substrate technologies and use a passive flow method to deliver fluids and reagents for assays. While numerous analogies between microfluidics and semiconductor technologies have been espoused, the physical differences between the fluid dynamics and electrical current are significant which suggests that current trends in physical design for microfluidics must change course in order to be of practical use to designers. Within this framework, a methodology is introduced to address design automation such as dynamically placing and routing microfluidic components in a non-discrete design space while avoiding invalid design layouts, accounting for fluid volume usage, surface area utilization, and the timing required to perform specified biological assays and also optimizing device parameters, enabling researchers to focus on the science and thereby accelerating the development of new, low-resource paper microfluidic devices for a developing world.

The principle of compositionality is axiomatically assumed in many modern syntacticand semantic research programs, and is integral to the interface of syntax and semantics.Compositionality is the idea that the meaning of a complex linguistic construction generallyought to be derivable based on the meanings of its components and how they are combined.(Frege (1988) Pelletier (1994), [2016], Janssen (2012)) Ellipsis, the systematic inference ofmeaning in the absence of an explicit linguistic string, is a phenomenon of considerableinterest on the syntax-semantics interface precisely because it is one area where the typicalassumptions of compositionality break down. As a result, it is an area where this and otherconstraints of syntactic and semantic theory can be examined and tested. This dissertationarose from a formal research program concerning certain elliptical constructions in English,gaps and strips, which are discussed in more detail and investigated from a variety ofmethodological perspectives throughout the rest of this book.In their 2016 paper, Kubota and Levine put forth an analysis of gapping (G) (1-a) inEnglish in the Type-Logical Grammar (TLG) framework. A subsequent paper, Puthawala(2018) extended this analysis to cover stripping phenomena (S) (1-b). The two analysestaken together license combinations of those constructions such as strip-gaps (SG) (1-c)and gap-strips (GS) (1-d). However examples of these complex constructions, particularlysingle-speaker utterances, are rare and difcult to fnd in corpora, and proved unacceptableto some speakers. These facts gave rise to two questions: First, are strip-gaps and gap-stripswell-formed in English? If evidence suggests that they are, it would validate the predictionsmade by the combined Kubota & Levine and Puthawala (KLP) analysis. If they aren’t,then it provides a straightforward way to challenge the analysis. Supposing that there is reason to believe that strip-gaps and gap-strips are licit for English speakers, then thesecond question then becomes why are they unacceptable to some people?(1) a. Mary picked carrots and Ryan tomatoes.b. Mary picked carrots and Ryan too.c. Mary picked carrots and Ryan too and David tomatoes.d. Mary picked carrots and Ryan tomatoes and David too.In this dissertation, I investigate these constructions through psycholinguistic experiments and the lens of the adaptive control of thought-rational (ACT-R) Parsing framework(Anderson (2005), Lewis and Vasishth (2005), Rasmussen and Schuler (2018) inter alia),which models the domain-general function of cognition and memory and yields output predictions of difculty and reading time. These sentences contain structures and features thatare often taken in the mainstream literature on parsing to be predictive of problems due toconstraints on working memory, predictability, repair, or ambiguity resolution, and I showthat the processing difculty could therefore provide an alternative explanation for the rarity and potential unacceptability of otherwise well-formed constructions, in support of thisexplanation, I develop a parsing algorithm to interface between the competence grammarof TLG and the processing model of ACT-R. This Type-Logical ACT-R parser representsan advancement in formal syntactic rigor over the standardly referenced models in the literature while being compatible with fully parallel models of human sentence processing,surprisal theory to model garden path efects (e.g. Levy (2008)), and neuroanatomicallyplausible models such as distributed associative memory (Rasmussen and Schuler (2018))

Action entropy active sensing represents huge leaps forward in the field of active sensing.Previous papers have found that while action entropy increases the accuracy of thesimulation, it also increases the time required to calculate sensing actions. The time costcan be reduced by implementing parallel processing that splits the computations overmultiple cores. This can be done on a local machine and via implementation of the cloud.This paper focuses on implementing a detection and recovery mechanism for a potentialcloud crash. The communication is facilitated through Docker and OpenVPN, using theROS client-service architecture. This paper adjusts the methodology of the existingmodel to make it more robust to failures. The simulation results indicate that the time torecovery is feasible, and that this new structure increases the reliability of the existingmodel

This paper aims to present a study on developing a program that assists nonprofit organizations in determining the ideal location for building their facilities based on community needs, thus maximizing their potential for success. The study highlights the importance of location in the success of nonprofit organizations, and the challenges they face in identifying suitable areas for their operations. The paper reviews existing literature on nonprofit organizations, location analysis, and data analysis techniques, and proposes a methodology for developing the program. The methodology involves data collection and analysis, and machine learning algorithms to predict community needs. The program provides a user-friendly interface for nonprofit organizations to access and analyze the data and offers recommendations for suitable locations based on their criteria. The study concludes that the proposed program can be a valuable tool for nonprofit organizations to make informed decisions about their location and maximize their potential for success in serving their communities.

Opportunities to learn computer science (CS) in K-12 have greatly expanded in the last decade. Driving this expansion is the idea that to be engaged and productive citizens, it is not enough for students to use technology; they must experience technology as producers and creators. Learning computer programming is one way for students to engage in creativity, an important twenty-first century skill often hard to find in schools. As the opportunity to learn CS grows in schools, teachers, many of whom are new to CS themselves, find themselves leading students in creative work in computer programming. This dissertation presents three perspectives on the pedagogy of creative work in high school computer programming. As context, I focus on teachers of AP CS Principles, an Advanced Placement introductory CS course offered in high schools across the U.S. Importantly, the course frames CS as a creative field and includes as part of its AP assessment the Create Task, a portfolio task for which students submit a program of their own creation. Using in-depth qualitative interviews of 35 AP CS Principles teachers and additional classroom observation data, I present three descriptive papers examining the pedagogy of creative work as students learn to program. In the first paper, I use teachers’ definitions of creativity to examine how they make sense of creativity within the domain of computer programming. I find that teachers identify ways that students display both expressive and technical creativity, both of which teachers connect strongly to the teaching and learning of computer programming. In the second paper, I analyze teachers’ descriptions of creative tasks to examine different framings of creative work in the classroom. I identify six framings: replication, re-creation, application, incrementation, synthesis, and self-direction. Finally, in the third paper, I examine the instructional practices teachers use when engaging students in creative work. I find that teachers contrast their teaching with traditional ideas of schooling. They describe evolving pedagogical practices that highlight student ownership, that distribute expertise horizontally, and that attend to students’ affective experience as they build identities as creators in CS.

The widespread expansion of Electric Vehicles (EV) throughout the world creates a requirement for charging stations. While Cybersecurity research is rapidly expanding in the field of Electric Vehicle Infrastructure, efforts are impacted by the availability of testing platforms. This paper presents a solution called the “Open Charge Point Protocol (OCPP) Cyber Range.” Its purpose is to conduct Cybersecurity research against vulnerabilities in the OCPP v1.6 protocol. The OCPP Cyber Range can be used to enable current or future research and to train operators and system managers of Electric Charge Vehicle Supply Equipment (EVSE). This paper demonstrates this solution using three attack types, Denial of Service, Machine-in-the-Middle, and Log4shell.

This paper introduces Prio+, a privacy-preserving system for the collection of aggregate statistics, with the same model and goals in mind as the original and highly influential Prio paper by Henry Corrigan-Gibbs and Dan Boneh (NSDI 2017). As in the original Prio, each client holds a private data value (e.g. number of visits to a particular website) and a small set of servers privately compute statistical functions over the set of client values (e.g. the average number of visits). To achieve security against faulty or malicious clients, unlike Prio, Prio+ clients use Boolean secret-sharing instead of zero-knowledge proofs to convince servers that their data is of the correct form and Prio+ servers execute a share conversion protocol as needed in order to properly compute over client data. This allows us to ensure that clients' data is properly formatted essentially for free, and the work shifts to novel share-conversion protocols between servers, where some care is needed to make it efficient. Our overall approach is simpler than Prio and our Prio+ strategy reduces the client's computational burden by at least two orders of magnitude (or more depending on the statistic) while keeping server costs comparable to Prio. Prio+ permits computation of exactly the same wide range of complex statistics as the original Prio protocol, including high-dimensional linear regression over private values held by clients. We report detailed benchmarks of our Prio+ implementation and compare these to both the original Go implementation of Prio and the Mozilla implementation of Prio. Our Prio+ software is open-source and released with the same license as Prio.

There are three main research conducted in this paper, including using deep learning methods with vision-based technologies on Structural Damage Detection (SDD), Structural Health Monitoring (SHM) and progressive collapse study. During the learning and improvement process, many goals of automation in SDD and SHM have been achieved, although there will be a large room for further improvement and development on these studies. In progressive collapse study, remote sensing technologies and data fusion are applied on a field experiment of a real building at the Central Campus of the Ohio State University. The major contributions of this paper are shown as follows:A few comprehensive experimental studies for automated SDD in extreme events using deep learning methods for processing 2D images. In the first study, a 152-layer Residual Network (ResNet) is utilized to identify multiple classes in eight SDD tasks, which include identification of scene levels, damage levels, material types, etc. The proposed ResNet achieved high accuracy for each task while the positions of the damage are not identifiable. In the second study, the existing ResNet and a segmentation network (U-Net) are combined into a new pipeline, cascaded networks, for categorizing and locating structural damage. The results show that the accuracy of damage detection is significantly improved compared to only using a segmentation network. In the third and fourth studies, end-to-end networks are developed and tested as a new solution to directly detect cracks and spalling in the image collections of recent large earthquakes. One of the proposed networks can achieve an accuracy above $67.6\%$ for all tested images at various scales and resolutions, and shows its robustness for these human-free detection tasks.Studies are conducted with a pipeline to automatically track and measure displacements and vibrations of structures or structural components in laboratory and field experiments. This novel framework that uses computer vision and deep learning methods to mimic human vision system for the dynamic performance assessment of in-service infrastructure with various camera placements. On one hand, the static deformations of cylinders and small-scale reinforced concrete beams in the laboratory tests are captured and measured by the proposed framework at first. Then two shaking table tests in the lab are utilized to assess the dynamic performance of the simulated structures. On the other hand, several bridges, including pedestrian, railway, and traffic bridges, are tested for their dynamic performance in field experiments with different camera placements: remote, structure-mounted, and drone-mounted cameras. To remove systematic motions of cameras and to capture the fundamental frequency of these tested structures, two techniques, displacement subtraction and frequency subtraction, are applied. To better understanding of practical applications, critical parameters for camera settings and data processing techniques, such as video frame-rates, window size and locations, and sampling rates on visual data are studied. It shows that not only the vibrations and the frequencies of the simulated structures (i.e., in lab tests) or the in-service structures (i.e., in filed experiments), but also the static deformation of structures or structural components, can be tracked and measured accurately by our proposed framework.The performance of a building in progressive collapse study is monitored and analyzed with the methods developed in previous studies. On one hand, we applied the proposed methods to detect structural damage on this reinforced concrete (RC) structure that was tested, including visual data from cell phones, inside and outside cameras, and drones. The experiments indicate that these solutions for automatic detection of structural damage using deep learning methods are feasible and promising. On the other hand, the proposed methods to measure the deformations and vibrations are utilized to process visual data from outside and inside cameras, and drones. Data from different sources are fused for capturing the performance of the structural elements under the scenarios of the sudden loss of the columns and slabs. The data fusion technique is useful to investigate the characteristics of this framed structure, especially when the traditional sensors such as strain gauges and Linear Variable Displacement Transducers (LVDTs) didn't work in the experiments. Since visual data acquisition and preparation, and techniques of data processing have been inclusively researched for real applications with deep learning, many experimental studies on SDD and SHM are carried out and promising results are obtained in this paper. In addition, the vibration-based technologies or traditional sensors are used as the reference. Our goal is to meet the needs for automatic detection of structural damage and accurate measurements for the performance of the structures or structural elements, thus, the efficiency and effectiveness of these frameworks are tested and analyzed. In summary, these research in this paper indicate that vision-based technologies with deep learning can be applied well on structural engineering domain, and facilitate structural engineers' job by providing reliable data and credential results. The methodologies developed in this paper can fill the gap of research and engineering applications in the future.

My dissertation comprises (1) the development of a machine learning framework that combines verbal and visual features together, models the intricate web of relationships between them, and extracts visual semantics, and (2) the application of a deep learning and a transfer learning framework to extract emotions from social media posts. This dissertation consists of three papers as follows. The first paper introduces a machine-learning visual framing analysis to examine the visual and verbal patterns of online news reporting and explore image-text relations in news stories. The second paper presents a machine-learning multimodal framing analysis to integrate the various types of data (e.g., image, text, and metadata) simultaneously and extract the semantic meaning from them together. The third paper is an application of a deep learning and a transfer learning to show the power of Twitter in providing fine-grained measures of real-time emotions and thereby offer a comprehensive overview of the role of emotions in voting participation. My dissertation can take into account various types of data simultaneously and extract politically meaningful semantics using computer vision, NLP, graph theory, high-dimensional statistics, and transfer learning.

The evolution of the Web over the last three decades has led to a deluge of scientific andnews articles on the Internet. Harnessing these publications in different fields of study iscritical to effective end user information consumption. Similarly, in the domain of healthcare, one of the key challenges with the adoption of Electronic Health Records (EHRs) forclinical practice has been the tremendous amount of clinical notes generated that can besummarized without which clinical decision making and communication will be inefficientand costly. In spite of the rapid advances in information retrieval and deep learning techniques towards abstractive document summarization, the results of these efforts continueto resemble extractive summaries, achieving promising results predominantly on lexicalmetrics but performing poorly on semantic metrics. Thus, abstractive summarization thatis driven by intrinsic and extrinsic semantics of documents is not adequately explored. Resources that can be used for generating semantics-driven abstractive summaries include:• Abstracts of multiple scientific articles published in a given technical field of study togenerate an abstractive summary for topically-related abstracts within the field, thusreducing the load of having to read semantically duplicate abstracts on a given topic.• Citation contexts from different authoritative papers citing a reference paper can beused to generate utility-oriented abstractive summary for a scientific article.• Biomedical articles and the named entities characterizing the biomedical articlesalong with background knowledge bases to generate entity and fact-aware abstractivesummaries.• Clinical notes of patients and clinical knowledge bases for abstractive clinical textsummarization using knowledge-driven multi-objective optimization.In this dissertation, we develop semantics-driven abstractive models based on intra-document and inter-document semantic analyses along with facts of named entities retrieved from domain-specific knowledge bases to produce summaries. Concretely, we propose a sequence of frameworks leveraging semantics at various granularity (e.g., word,sentence, document, topic, citations, and named entities) levels, by utilizing external resources. The proposed frameworks have been applied to a range of tasks including1. Abstractive summarization of topic-centric multi-document scientific articles andnews articles.2. Abstractive summarization of scientific articles using crowd-sourced citation contexts.3. Abstractive summarization of biomedical articles clustered based on entity-relatedness.4. Abstractive summarization of clinical notes of patients with heart failure and ChestX-Rays recordings.The proposed approaches achieve impressive performance in terms of preserving semantics in abstractive summarization while paraphrasing. For summarization of topic-centric multiple scientific/news articles, we propose a three-stage approach where abstractsof scientific articles or news articles are clustered based on their topical similarity determined from topics generated using Latent Dirichlet Allocation (LDA), followed by extractive phase and abstractive phase. Then, in the next stage, we focus on abstractive summarization of biomedical literature where we leverage named entities in biomedical articles to1) cluster related articles; and 2) leverage the named entities towards guiding abstractivesummarization. Finally, in the last stage, we turn to external resources such as citationcontexts pointing to a scientific article to generate a comprehensive and utility-centric abstractive summary of a scientific article, domain-specific knowledge bases to fill gaps ininformation about entities in a biomedical article to summarize and clinical notes to guide abstractive summarization of clinical text. Thus, the bottom-up progression of exploringsemantics towards abstractive summarization in this dissertation starts with (i) SemanticAnalysis of Latent Topics; builds on (ii) Internal and External Knowledge-I (gleaned fromabstracts and Citation Contexts); and extends it to make it comprehensive using (iii) Internal and External Knowledge-II (Named Entities and Knowledge Bases).

Out-of-distribution (OOD) detection aims to identify test examples that do not belong to the training distribution and are thus unlikely to be predicted reliably. Despite a plethora of existing works, most of them focused only on the scenario where OOD examples come from semantic shift (e.g., unseen categories), ignoring other possible causes (e.g., covariate shift). In this paper, we present a novel, unifying framework to study OOD detection in a broader scope. Instead of detecting OOD examples from a particular cause, we propose to detect examples that a deployed machine learning model (e.g., an image classifier) is unable to predict correctly. That is, whether a test example should be detected and rejected or not is “model-specific”. We show that this framework unifies the detection of OOD examples caused by semantic shift and covariate shift, and closely addresses the concern of applying a machine learning model to uncontrolled environments. We provide an extensive analysis that involves a variety of models (e.g., different architectures and training strategies), sources of OOD examples, and OOD detection approaches, and reveal several insights into improving and understanding OOD detection in uncontrolled environments.

Blockchain is a decentralized, peer-to-peer (P2P) network that is becoming more popular because of its many benefits, such as reliability, scalability, and privacy over centralized networks. This technology gained traction in cryptocurrencies originally then it spread into the industry. But centralized networks still outperform decentralized networks in broadcast speed. Blockchain transactions have volatile latencies ranging from minutes to months, depending on the network traffic, while centralized transactions take milliseconds to complete. As network transactions increase, blockchain messages will reach their destinations slower. In this paper, we attempt to solve this problem in a decentralized manner by optimizing the P2P network's topology and making each unconnected node choose its best local connections. In an environment where network nodes trust each other and cannot connect to malicious nodes, these networks can decrease latency by creating node embeddings and predicting a node's best local connections with our AI model that consists of Multi-armed Bandits, Graph Convolutional Neural Networks, and Reinforcement Learning.

Radio frequency (RF) refers to a band of electromagnetic waves ranging from 20 kHz to 200 GHz. RF has been widely used for wireless communication networks, such as WiFi, as well as for industrial and household appliances like microwave ovens. With the advancement of technologies and the densification of radio nodes in the limited spectrum, different protocols and standards have been used to support seamless and efficient wireless sensing and communication among radios. Radio signals get refected, refracted, and attenuated as they interact with the surrounding environ- ment. Such interactions can be measured to introduce innovative radio frequency sensing techniques. Food analytics and estimation of food nutrients have been widely used in recent years to monitor and control food intake and calorie consumption by individuals. Microwave ovens have been widely used in recent years to heat and cook food quickly and efciently. The microwave generates electromagnetic waves in the RF spectrum, at 2.4 GHz frequency to heat polar molecules in the food. This radi- ation leaks out through the front door panel of the microwave oven. However, using the microwave, it has been a challenge to estimate food composition and how long it needs to be heated to a particular temperature.This thesis presents systems and studies that explore novel radio-frequency (RF) sensing applications to solve these challenges using microwave oven.Users estimate the time to heat food in a microwave oven by prior knowledge or by trial and error process. However, this often results in the food being over-heated or under-heated, destroying the nutrients. The state-of-the-art approaches require hardware installation (thermal camera, temperature sensors) on the microwave oven, which is not cost-efcient and cannot be installed by users easily. In this thesis, we present RFTemp, a system that can monitor microwave oven leakage to estimate the temperature of the food that is being heated and thus estimate the accurate time when the food has reached the targeted temperature.On the other hand, user estimate the food nutrient composition either using some lookup information of each of the ingredients of the food. This is often time consuming and not trivial for complex foods. In this paper, we present WiNE, a system that can monitor microwave oven leakage to estimate the food nutrient (carbohydrate, fat, protein, water) composition and thus can measure the calorie intake of the foodThroughout the thesis, we implemented our systems using software-defned radios (WARP V3) and performed extensive experiments in 2.45 GHz band. Our results show that our proposed systems introduce a novel RF sensing domain for microwave ovens. The last part of the thesis summarizes the contributions of the research and presents the conclusion of the study. It also suggests potential areas for future work in order to further enhance their performance.

Deep Learning has achieved state-of-the-art performance in several artificial intelligence tasks like object recognition, speech recognition, machine translation, and summarization. Deep learning is a subset of machine learning that learns multiple levels of data representation using Neural Networks (NNs). The rise of deep learning can be attributed to the presence of large datasets and computation power. Large-scale Deep Neural Networks (DNNs) can provide state-of-the-art performance by learning complex relationships, enabling them to push the boundaries in artificial intelligence. However, training such large-scale DNNs is a compute-intensive task as it can have billions of parameters, which increases both the memory and computational requirement of DNN training. Hence, distributed DNN training has become the default approach to train large-scale DNNs like AmoebaNet, GPT3, and T5.Broadly, the DNN training pipeline can be divided into multiple phases: 1) Data Loading and Data Augmentation, 2) Forward/Backward Pass, and 3) Model Validation. Traditionally, these phases are executed sequentially on a single CPU or GPU due to a lack of additional resources. Multiple processing elements can be used to parallelize the computation in each phase and reduce the overall training time. In this dissertation, we propose novel parallelization strategies for distributed DNN training to alleviate bottlenecks in different phases of DNN training and parallelize the computation across multiple processing elements. Novel parallelization strategies are required to efficiently distribute the work among multiple processing elements and reduce communication overhead, as naive parallelization strategies may not give performance benefits when distributing work among multiple processing elements because of high communication overhead. Therefore, we need novel parallelization strategies designed to distribute the work while keeping the communication overhead low. There are several challenges in the existing DNN training pipeline. Data loading/augmentation and model validation can be up to 20% of the overall training time, making training large-scale DNNs time-consuming. Therefore, we propose a new parallelization scheme that uses the computing power of NVIDIA's recently released Data Processing Units (DPUs) to offload data loading and model validation phases and accelerate the performance of Data Parallelism.Forward and backward passes remain the most compute-intensive phase in the DNN training pipeline. Increasing the number of layers and parameters in DNNs to achieve better accuracy has become a common approach in deep learning. In the last couple of years, several DNNs like AmoebaNet, T5, and GPT3 have been proposed in the literature, pushing the boundary of the number of parameters and layers. However, computation and memory requirements also increase with the increase in the number of layers and parameters. Therefore, these models cannot be trained on a single processing element. Broadly, large-scale DNNs can be categorized into two categories: 1) In-core models (DNNs that fit inside the memory of a single processing element) and 2) Out-of-core models (DNNs that are too large to fit inside the memory of a single processing element). Technically, the in-core model can be trained on a single processing element, but the training time will be too high, which makes the training impossible. Therefore, we need a novel parallelization strategy that accelerates the training and uses the inherent parallelism of DNN architecture. Because of limited memory in modern accelerators like GPUs, several large-scale DNNs cannot be trained on a single processing element. Therefore, we need novel parallelization strategies to distribute the layers/neurons among multiple processing elements to reduce the memory requirement on each processing element. In this dissertation, we propose several novel parallelization strategies to alleviate current bottlenecks in different phases DNN training pipeline and reduce the overall training time. The key idea is that one should develop custom parallelization strategies for each DNN architecture type so that it can utilize the inherent parallelism and computation pattern to reduce the training time while keeping it generic and applicable to a large number of models in deep learning. We have developed several novel parallelization strategies like Data Sub-Graph Parallelism, Bi-Directional Parallelism, and Hybrid Five-Dimensional Parallelism that accelerate DNN training for in-core, out-of-core model, and out-of-core layer DNNs, respectively. Developed strategies are evaluated on large-scale GPU systems and are made available as public releases or published papers.

High-level synthesis (HLS) tools simplify the FPGA design processes by allowing users to express their designs in high-level languages such as C/C++ or OpenCL. In this way, users could focus on algorithmic optimization with less concern for the cycle-by-cycle details at the register-transfer level (RTL). However, FPGA development flows still have two major limitations that hinder the adoption of FPGAs:- Limited achievable frequency. There still exists a considerable gap between the quality-of-result (QoR) of an HLS-generated design and what is achievable by an RTL expert, especially the maximum operating frequency of the design. With the designs being scaled up in size, the final achievable frequency will be even lower. Unfortunately, a frequency degradation will directly lead to a proportional performance drop.- Prolonged compilation time. In the current FPGA CAD flow, the RTL generated by the HLS compiler will be passed to the traditional synthesis and implementation tool. Although the C-to-RTL compilation is relatively quick, the RTL-to-bitstream implementation process will take much longer. With the designs becoming increasingly complex and the FPGA devices larger, the compile time surges from hours to days. Such an overlong process will seriously limit the working efficiency of engineers, especially when compared to software compilation that only takes seconds or minutes.We observe that the existing FPGA CAD flows have not taken full advantage of HLS for further timing optimization and compilation reduction. Currently, the synthesis, placement, and routing tools are implemented and optimized to handle arbitrary RTL inputs. Those tools will adhere to the cycle-accurate behavior of the input design to ensure the correctness of the output. However, HLS-generated RTL is highly flexible and may tolerate additional pipeline registers without causing functional errors. Such latency-insensitive properties could significantly help the downstream compilation with timing closure. However, in the current toolchains, the HLS compilation is a standalone step, and the HLS-generated RTL will be treated in the same way as manually-written RTL by the logic synthesis tool. As a result, the information on pipeline flexibility in HLS designs will be lost, and the downstream physical implementation process cannot insert pipeline registers for timing closure. Based on this observation, we propose methods to co-optimize the HLS compilation and the physical design process, which will enable frequency improvement and speed up the hardware accelerator development process simultaneously. Different from the conventional compilation stacks that separate the HLS compilation from the downstream physical implementation process, we propose to bridge the gap between HLS and physical design organically. By facilitating placement and routing with the latency-insensitive information of HLS, and in turn by guiding the HLS compilation with the physical layout information, we could achieve significant improvement in QoR and reduction in compile time. Centered around this core idea, my thesis consists of three major parts. First, we explore how to improve the inherent timing quality of the RTL generated by HLS. Next, we couple HLS scheduling with coarse-grained floorplanning to improve the achievable frequency. Finally, we take one step further by partitioning the design for parallel placement and routing, then efficiently stitch them together without losing timing quality.First, the thesis addresses the timing-closure challenge by improving the inherent timing quality of the machine-generated RTL. This chapter studies the timing issues in a diverse set of realistic and complex FPGA HLS designs, including two of my previously-published accelerator designs for genome sequencing. We observe that in almost all cases, the frequency degradation is caused by the broadcast structures generated by the HLS compiler. We classify three major types of broadcasts and propose a set of effective yet easy-to-implement approaches. Our experimental results show that our methods can improve the maximum frequency of a set of nine representative HLS benchmarks by 53\% on average. In addition to optimizing the QoR of HLS by itself, the thesis further pushes up the final frequency by coupling HLS compilation with floorplanning. We propose AutoBridge, an automated framework that couples a coarse-grained floorplanning step with pipelining during HLS compilation. Since pipelining may introduce additional latency, we further present analysis and algorithms to ensure the added latency will not compromise the overall throughput. In our experiments with a total of 43 design configurations, we improve the average frequency from 147 MHz to 297 MHz (a 102% improvement) with no loss of throughput and a negligible change in resource utilization. Notably, in 16 experiments, we make the originally unroutable designs achieve 274 MHz on average. AutoBridge was recognized with the Best Paper Award in FPGA 2021.Finally, we take one step further to enable parallel physical implementation on top of our HLS-floorplan co-design methodology. We propose a split compilation approach based on the pipelining flexibility at the HLS level. The pipeline flexibility allows us to partition designs for parallel placement and routing without timing degradation. Our research produces RapidStream, a parallelized and physical-integrated compilation framework that takes in a latency-insensitive program in C/C++ and generates a fully placed and routed implementation. When tested on the AMD/Xilinx U280 FPGA, we observed a 5-7X compile time reduction and a 1.3X frequency increase. RapidStream was recognized with the Best Paper Award in FPGA 2022.In conclusion, my thesis targets two of the most challenging problems for modern EDA tools: timing closure and agile compilation. We first study the fanout optimization at the HLS level. Next, we explore the co-optimization of HLS and floorplanning, which has been used by at least eight other accelerator design projects. Finally, we enable the split compilation of HLS designs to reduce the compile time significantly. At the end of the thesis, we discuss future directions, including extending the methodology to support the compilation of RTL designs, multi-FPGA designs, and ASIC designs.

The world is overflowing with text. This ever-growing resource has the ability to capture thoughts, ideas, and understandings. One example is the scientific research paper which often contains a new discovery or details an in depth understanding---and new research papers are growing at an exponential rate, a rate that may be difficult for the human mind to keep up with. Additionally, the increase of collaboration across different knowledge domains requires an increased effort for a specialist of any one field to understand. These increases can lead to mistakes among researchers from missing important information that has already been documented. Similarly, the electronic health record is increasing rapidly as technology is integrating itself into the patient physician interaction. To be able to deliver information quickly and accurately to a physician can help ease the burden and lesson the mistakes that a primary care physician can make when dealing with the increasing pressure from seeing too many patients in too little time. Given the enormous amount of textual data, computational techniques must be developed that can effectively process the data. This work presents approaches that seek knowledge discovery from a large input of biomedical and scientific text. In the context of scientific research papers, we discuss how and why we need to automate the scientific method using a causal pipeline. Starting with the raw text of a scientific corpus we demonstrate the ability to improve scientific decision making in experiment planning and deductions. For the task of summarizing corpora, we introduce a new topic model that seeks to model topics off a pre-existing knowledge source. As we show empirically, our methods of extraction, connection, and summarization of relevant electronic text records results in knowledge discovery and new understandings.

Deep learning (DL) algorithms are prevalent in radiology as workflow assistants and as modality enhancements. Magnetic resonance imaging (MRI), computerized tomography (CT), diagnostic imaging (DI), ultrasound (US), and positron emission tomography (PET) are modalities that benefit from the DL algorithms and shorter exam times or greater image accuracy. Faster scan time is achieved by the signal to noise ratio (SNR). The distinction is that the technology can enhance images beyond the original resolution from the modality or shorten exam time and rebuild the image quality through SNR algorithms back to approximately the original standard of care (SOC) image. While artificial intelligence signal to noise ratio algorithms (AI-SNR) can enhance an image to a greater accuracy, shorter exam times are a measurable component of a return on investment (ROI) in calculating modality utilization. The algorithm-derived images may have visual variations that are not found on normally acquired original images. The research focused on DL-SNR images on a three-tesla magnetic resonance imaging (3.0T MRI) unit, a high resolution MRI deployment in the industry. The primary research question for this research study is: What are radiologists’ perceptions in regard to image quality and increased utilization due to vendor provided DL-SNR on 3.0T MRI? This will be an exploratory qualitative research study using detailed interviews with fellowship-trained radiologists that are using AI-SNR in 3.0T MRI and shortened exam time protocols. Fifteen interviews were conducted. The interview transcriptions were coded using ATLAS.ti to identify common themes and sub-themes in the radiologists’ perceptions of DL-SNR imaging. This paper assumes the reader has an adequate understanding of deep learning and radiology processes. The interviews included discussions on key elements on image quality, workflow, reimbursement, legal concerns, and radiologist workload. Issues were identified and potential solutions for optimizations were rendered.

Software models are used to analyze and understand the properties of the system, providing stakeholders with an overview of how the system should work before actually implementing it. Such models are usually created informally, such as drawing sketches on a whiteboard or paper, especially during the early design phase, because these methods foster communication and collaboration among stakeholders. However, these informal sketches must be formalized to be useful in later applications, such as analysis, code generation, and documentation. This formalization process is often tedious, error-prone, and time-consuming. In an effort to avoid recreating formal models from scratch, this thesis presents SkeMo, a sketch-based software modeling tool. SkeMo is built on a CNN-based image classifier using 3000 input sketches of class diagram components and integrated into the functionality of an existing web-based model editor, the Instructional Modeling Language (IML), with a newly implemented touch interface. SkeMo was evaluated using a ten-fold cross-validation to assess the image classifier and through a user study involving 20 participants to collect metrics and feedback. The results demonstrate the promising potential of sketch-based modeling as an intuitive and efficient modeling practice, allowing users to quickly and easily create models to design complex software systems.

Over the years, it has become increasingly common for developers to reuse open source code during software development. While this can save them time and effort in development, software developers tend to misuse or misunderstand the software licenses in free and open source software projects (FOSS). Software licenses dictate how software can be modified, reused, and distributed. In the case of FOSS, licenses generally fall under two main categories: permissive and restrictive. When a license violation occurs in a developer’s project, failure to address it can lead to legal issues such as being prevented from distributing their software or even monetary implications. When addressing a library with a violation, replacing it manually can often be time consuming especially when the library has many features used in the developer’s project. In this paper, we propose a semi-automated approach to assess license compatibility within a codebase and suggest replacement method calls from another library. We implement our tool as an Eclipse plugin and demonstrate our approach is capable of finding potential license violations accurately and able to successfully recommend alternatives in both artificial and open source Java projects from GitHub.

Going from the description of a model architecture in a figure to its implementation can be a fraught process. This work presents a markup language for specifying model architectures; an associated Python package used to convert models into a runnable format; utilities to import the model into PyTorch; a publicly available model repos- itory; and a tool to visualize the resulting models. This paper also provides a review of some popular machine learning architectures, examples of models created using the markup, and associated experiments. The system is named Agrippa * , determined through a bracket-poll tournament conducted by the author’s roommates. The language simplifies certain aspects of model development: parameters are named, explicit, and can be specified as being frozen or shared; models can be imported into different projects with few code changes; coherent parts of the architecture can be or- ganized into self-contained blocks; and parameter initialization techniques are explicit. The language syntax is simply XML. Models compiled using the Agrippa Python pack- age are converted into the ONNX format, a neural network interchange format sup- ported by a variety of machine learning frameworks. The web component of Agrippa can be found at http://agrippa.build , which con- tains the visualization tool, model repository, and documentation. † At this stage of development, the system has a number of limitations: models may not be compiled if they are larger than 2GB due to a formal limit imposed by the ONNX file format; not all operations available in ONNX are supported by the com- piler; the visualization tool does not yet support visual editing; and certain techniques, like dropout layers and batch normalization, require workarounds. There are a number of plausible ways each of these limitations may be addressed in the future.

With the rapid increase of social network spam, it's essential to empower users with the tools to detect the harmful spam effectively. However, existing tools cannot meet the requirements. In this paper, we propose and develop a live detection tool that can detect ham and spam text and images from social networks, this tool will be trained on user collected data (Image and Text) using different classifiers, where text and images are pre-processed and then passed onto the classifier that the user can choose, the user is then able to save the model and load it whenever they want to use a social network, where this tool will show the user a notification alerting them whether the post they are looking at is spam or ham before they even get the chance to read the text or look at the image, thus protecting them from clicking on malicious links that might harm their computer and steal their data. Evaluation results have demonstrated the effectiveness of our tool.

Diabetes is one of the highest causes of death in the world. Diabetes is caused when theblood glucose level is too high in the body. Gradually, high blood glucose leads to heartdisease, stroke, eye, and foot problems. To prevent the dreadful effects among people,early detection is required that would lead to proper medical treatment and change inlifestyle. Therefore, with the rise of machine learning we can predict if a patient hasdiabetes or not. Furthermore, we will integrate the trained model to a web applicationthat will connect the model to generate predictions in real-time considering factorsresponsible for diabetes like body mass index (BMI), age, insulin, etc. In this paper, we areusing the Pima Indian dataset that is originally from the National Institute of Diabetes,Digestive and Kidney Diseases for diabetes prediction model design using machinelearning. The proposed system in this paper is the Soft Voting ensemble classifier. Thealgorithm with the best accurate result was used in making predictions. This model wasdeployed to the web using flask (a python framework), it takes inputs from the user tomake predictions. This model is implemented using python programming language andflask (a web base framework) hosted in GCP. Soft Voting ensemble classifiers evenperform better than other classifiers with an accuracy of 91.55% which is quite promisingconsidering the other classification models in the literature for this problem.

The question, what spaces admit Locality Sensitive Hash families is of great importance to the field of data-analytical hashing (such spaces are called LSH-able). This question has been restated and examined with each update to the definition. Particularly important to this work is the definition proposed by Charikar defined over similarity measures that is used in the seminal paper to give the first formal proofs about LSHability. The study of Collision Probability Functions(CPFs) has led to the relaxation of some of the requirements. This thesis focuses on the ideas presented in the papers on Distance Sensitive Hashing and Asymmetric Hashing. In particular this thesis will take the idea that LSH families only need to have well-defined CPFs to be studied generally and explore the benefits of such a view of Hashing while relaxing the traditional LSH requirements further.This thesis proposes a novel definition of hashing by analyzing a recently developed concept in the field of hashing known as non-transitive hashing. This thesis will expand theoretically on the notion of non-transitive hashing in order to develop a more general framework for LSH. This thesis shows that the original argument for non-LSHability of a space introduced by Charikar no longer applies when non-transitive hash functions are permitted. Some similarity measures that seem unable to be hashed per the triangle inequality argument of Charikar can be when the notion of hash collisions is expanded non-transitively.This thesis will justify the use of non-transitive hashing by examining known applications that violate transitivity. The fundamental concepts of Or-Amplification and hash signatures can and often do violate transitivity. To accommodate this change this thesis also suggests a change in notation for talking about non-transitive hash collisions, namely using the traditional symbol for mathematical relation,~, instead of =. The relation might depend on the hash scheme and its complexity factored into the analysis of the algorithms one might use it for. In order to well define one particularly large class of non-transitive hashing applications, this thesis proposes the Boolean Hash Collision Relation in order to give a consistent theoretical understanding of what constitutes hash families built via Amplification of other families.Using this non-transitive approach to hash function collision, this thesis will show that certain spaces are strongly LSHable even though the corresponding dissimilarity violates the triangle inequality. This thesis will explore other similarity measures that might be LSHable or approximately hashed using non-transitive methods. This thesis applies those ideas to provide new classes of non-transitive hash families with asymmetric properties introduced in this thesis by way of NOT-amplification. Together with the familiar OR-Amplification and AND-Amplification, this thesis introduces the Boolean-collision relation to generalize the notion of Amplification and describe a large class of non-transitive hash families. Using this relation for hash comparison this thesis will provide an approximate hash of Sørensen Coefficient and Sokal-Sneath using non-transitive hashing.

IoT (Internet of Things) frameworks are designed to simplify the development process of IoT devices and applications. However, like any technology, they also have some problems that need to be addressed. Some of the common problems associated with IoT frameworks include:• Security Concerns: IoT devices collect and transmit sensitive data, making them a prime target for cyberattacks. Many IoT frameworks lack built-in security features, making it easier for hackers to access the data.• Interoperability Issues: IoT devices from different manufacturers may use different protocols and standards, making it difficult for them to communicate with each other.• Scalability Challenges: IoT systems can involve large numbers of devices and applications, and the frameworks used to develop these systems need to be scalable. This requires significant computing power and data storage capacity, which can be a challenge for some IoT frameworks.• Complexity: IoT development can be complex, as it involves multiple components and technologies, such as cloud computing, big data analytics, and edge computing. The frameworks used to develop IoT systems need to be able to integrate these components and simplify the development process.• Cost: Developing and deploying IoT systems can be expensive, as it requires specialized hardware, software, and infrastructure. The cost of IoT frameworks can also be a barrier to adoption for some organizations.Just like commercial IoT frameworks, Department of Defense (DoD) unattended ground sensors (UGSs) come in many shapes and sizes and often have many different components. These components range from infrared cameras and magnetometers to communications equipment and embedded computers. The combination of these different components requires sophisticated software that is often time consuming to develop and difficult to reuse and have the same problems like commercial IoT systems.The US Army Research Laboratory (ARL) and the Defense Intelligence Agency funded the Open Standards for Unattended Sensors (OSUS) to address solving these problems.3 The OSUS program, formerly named Terra Harvest, was launched in 2009 to develop an open, integrated battlefield unattended ground sensors architecture that ensures interoperability among disparate UGS components and systems.The University of Dayton Research Institute (UDRI) software system group (SSG) was chosen to work with the ARL to create a reference implementation of the OSUS standard. ARL along with UDRI went to various field-testing and demonstrations to test out OSUS in real-life simulations, integrating various vendors sensor, controllers, radios, using the OSUS simple plug-in-play architecture showing a solution that overcame the typical problems of UGS/IoT systems.After years of field success, the US Army reclassified the technology, allowing it to be used by the public. This has led it to become the standard for various US agencies and The Five Eyes Alliance (FVEYs) intelligence communities. At the FVEY The Technical Cooperation Program’s (TTCP) Contested Urban Environment Strategic Challenge (CUE17, 18, 19, and 20 for short), OSUS was the only middleware layer for all of the events. The author led the team to integrate all the various worldwide sensors/systems using OSUS. OSUS has shown through all these demonstrations that it is able to integrate world-wide sensors/systems into a common framework, eliminating most of the problems with UGS/IoT systems as stated above.In this thesis, I will attempt to show that OSUS offers a solid framework to use for IoT. The paper will provide a background of what IoT is, along with the architecture of OSUS. It will then show how one can create a plugin for OSUS and lastly, how to deploy/test these plugins.

Students with Autism Spectrum Disorder (ASD) face challenges accessing traditional educational curricula. Technology has revolutionized classroom instruction, yet access remains limited for Students with ASD. This project seeks to design an inclusive computing curriculum for students with ASD based on Creative Computing Curriculum. The proposed curriculum is customized to address the unique needs of students with ASD, using visual aids, simplified language, instructional videos, and hands-on project-based learning activities that promote creativity and problem-solving abilities. By drawing upon creative computing principles, this curriculum strives to offer an engaging and accessible learning experience for Students with ASD. This paper details its development process, outlining strategies for adapting the Creative Computing Curriculum specifically for their needs. By creating an inclusive educational system we hope to promote equity and accessibility for all learners.

This paper describes the process of development and evaluation of an open educational resource (OER) e-module on the Central Limit Theorem written for an Introductory Statistics college-level course. The purpose of this project is two-fold. First, the e-module bridges the knowledge gap between introductory topics and Hypothesis Testing – one of the most challenging concepts in Statistics. Second, the project focuses on developing tools that allow instructors to analyze the effectiveness of the module and reveal student patterns of interaction with the platform. The overall goal of the project is to improve the quality of open educational resources, provide students/instructors with additional study materials in response to rising cost for textbooks and higher education, and provide more data for further research on student behavior while interacting with e-textbooks. The interactive e-module was developed using LaTeX markup language and Overleaf editor, uploaded to the XIMERA platform and tested on two sections of MTH 140, a college-level Statistics course. Once the experiment has been performed and the data collected, the results were analyzed using Python programming language. As a result of the study, some tools for analysis of user data have been developed, and an OER has been created.

Algorithms are used by social media platforms to gather information on users to better suggest updated content to maximize the time spent on their online platforms. The information gathered from the users (browsing history, search history, engaged posts, engaged channels, etc.) is analyzed by these algorithms and used to predict what content and channels to best offer the users for further engagement. Information such as liking patterns, following tendencies, content engagement behavior and more are fed to these algorithms to personalize the user’s experience. The goal of this paper is to compare the effectiveness of the algorithms’ ability to profile and induce users, in this case employed by Instagram, Twitter, and YouTube by tracking several variables, including interaction time and interaction rates with posts, channels, likes, and follows. The results of this study show the performance of each platform algorithm over weeks of observation and how they stack in terms of engagement effectiveness.

Image super-resolution (SR) is a classic problem in image restoration, which aims to reconstruct a high-resolution (HR) image with realistic details from a low-resolution (LR) image. The challenge lies in the fact that SR models need to ‘imagine’ details that may not exist in the LR image. Although recent state-of-art super-resolution models have made great improvements in performance, the models we have found are trained for open domain SR use. Our research addresses the lack of high-quality architectural images on the internet, including popular web mapping platforms like Google Maps 360◦ Streetview and Bing Streetside by proposing a super-resolution model refined for architectural images. We use a domain-specific dataset of architectural images of building facades collected by our team to train our model in order to analyze the performance of a specialized model over general super-resolution models. We modify the model architecture to generate better results in terms of perceptual realism, sharpness and fine details. In this paper, we introduce a super-resolution model specialized in architectural images to fill a void in the state-of-art SR model offerings.

Multivariate networks (henceforth, graphs) represent entities (vertices or nodes), their relationships to each other (edges), and manifest or derived data about both (attributes). Graphs easily map onto real-world entities and relationships. Therefore the visual analysis of such provide a valuable proxy for the analysis of concrete entities. Because graphs are so effective in modeling, they are ubiquitous: graphs are used in the design and construction of complex circuitry, GIS (Geographic Information Systems), database modeling, social networking, and neuroscience, among other fields. Graph visualization is challenging, and domain-specific requirements further complicate the production of intelligible visual representations. The datasets underlying modern graphs have grown exponentially, complicating visualization tasks, whether static, such as in a printed scientific paper or poster, or dynamic, where the graph, the visualization, or both evolve. Exploratory visualization of dynamical graphs requires the visualizations to evolve due to externally-generated events in realtime while preserving the contents of the analyst’s “mental map.” This thesis presents a novel, modular, composable Javascript API (Application Programming Interface) for organizing and applying the algorithmic operations and data structures required at each step of visualizing large, arbitrarily complex graphs while allowing the extensions and constraints required by specific application domains. To do so, we abstract quintessential graph layout tasks into EGOs (Exploratory Graph Operations), design an API exposing these operations, and demonstrate the synthesis of well-documented layout algorithms. Finally, we introduce a reference implementation in Javascript and WebGPU (GPU Computing for the Web) to visualize complex graphs in realtime and facilitate future research and development of novel approaches to these problems.

The origins of modern cryptography can be traced back to several milestones: Shannon's work on the perfect secrecy of symmetric encryption, the advent of public-key cryptography through the Diffie–Hellman key exchange, and the invention of the RSA public-key encryption scheme. Following these breakthroughs, the introduction of provable security laid out the methodological groundwork for arguing the security of cryptographic protocols, by relating their security to hard mathematical problems: if an attacker is able to invalidate the security of the scheme, then it must also be able to efficiently solve a problem for which no known efficient algorithm exists. The hardness of the problem with respect to specific parameters is formally framed as a cryptographic assumption. Over an expansive span of time, algorithmic research into these hard problems has the goal of investigating their properties and establishing various relations and hierarchies between them. In turn, this ever-changing complexity-theoretical landscape motivates the design of schemes that rely on either weaker assumptions or new assumptions altogether, offering alternatives in the eventuality of algorithmic breakthroughs that would invalidate previous solutions. Several years after the first public-key encryption was proposed, modern cryptographic research has expanded well beyond the confines of symmetric and public-key encryption. One of these developments is zero-knowledge proof systems for NP, in which a prover convinces a verifier that a problem instance belongs to a language included in NP, while at the same time avoiding the leakage of any information about the witnesses of the instance (beyond the knowledge that at least one such witness exists). In this thesis, we are mainly focused on building schemes that are round-efficient, with an interaction that is limited to only one or two messages: 1. we provide new constructions of non-interactive zero-knowledge proofs (NIZKs), where the entire interaction consists of one message from prover to verifier. 2. we build new ZAPs, which are a special case of two-message zero-knowledge proofs: the verifier initiates with the first message (which is a uniformly random string), and the second message must be the reply of the prover. Moreover, the verifier message should be securely reusable for potentially different statements for an arbitrary polynomial number of times. Since fully-fledged zero-knowledge is impossible in less than three rounds, only witness indistinguishability is required for ZAPs: honest proofs computed using different witnesses must be indistinguishable. Over the past two decades, NIZKs and ZAPs emerged as a crucial tool in the design of more complex protocols, owing that to both their round efficiency and their usefulness in arguing about computation. Due to the high expressivity of NP languages, these protocols allow to prove for instance that an output corresponds to a faithful execution of some polynomial-time program. This property has been leveraged to build complex cryptographic primitives, such as ring signatures, group signatures or delegatable anonymous credentials. We describe novel constructions whose security depends on assumptions for which no previous result existed, following the general cryptographic approach of being able to resort to alternatives should some of the previous assumptions be invalidated by algorithmic breakthroughs. One of our constructions relies on a standard cryptographic assumption that was not previously known to imply statistical ZAPs, while the other results are based on well-studied, but less standard assumptions. Relying on all these previously untackled assumptions requires devising new techniques along the way, therefore enlarging our knowledge about what properties are required from our assumptions in order to leverage them into constructions of ZAPs or NIZKs. Moreover, our results also lead to new relations between the cryptographic assumptions that we consider. Our first result concerns the relationship of zero-knowledge with indistinguishability obfuscation (IO). Informally, IO is a powerful assumption which allows to obfuscate any program in such a way that its functionality is hidden, but the program can still be executed. NIZKs and IO both offer different security guarantees for computation, and our work establishes a connection between these distinct notions. Moreover, our result also helps link IO with a family of algebraic assumptions. In more detail, we construct a NIZK with a dual-mode property, based on the subexponential security of indistinguishability obfuscation IO, and one-way functions, along with the polynomial security of an algebraic object with a lossy property (such as lossy encryption). The dual-mode property allows the common reference string to exist in two computationally-indistinguishable modes, requiring that one ensures statistical soundness and the other statistical zero-knowledge. Our construction has further implications, and in conjunction with previously known results, leads to new relations between IO and algebraic assumptions such as DDH, bilinear SDDH or multilinear k-MDDH. Our results are encompassed in a work published at Asiacrypt 2019. In our second result, we build a NIZK in cyclic groups without requiring the use of pairings. Pairings are a specific type of bilinear map, and have been successfully used to build NIZKs. This comes however at the cost of assuming more algebraic structure, and NIZKs in pairing-free cyclic groups have remained an important open problem until recently. More explicitly, we build a NIZK based on polynomially-hard Computational Diffie-Hellman (CDH), along with the 2^{-c\lambda}-OWKDM assumption, for constant c=3/4 (the constant is improved to 1/2 in a subsequent result). Defined over cyclic groups of size 2^\lambda, the 2^{-c\lambda}-OWKDM assumption is that efficient adversaries against the key-dependent message one-wayness of ElGamal are limited to an advantage of poly(\lambda)/2^{c\lambda}, where f is any efficient function f. The key-dependent message one-wayness of ElGamal means that given an ElGamal encryption of f(k), where k is the ElGamal key and f is an efficient function, the adversary cannot recover f(k). This result corresponds to our work published at Eurocrypt 2020. Finally, we design statistical two-message public-coin witness-indistinguishable proof systems (statistical ZAPs) for NP. Until a recent breakthrough result, statistical ZAPs in cyclic groups have remained elusive, even in the presence of pairings. Our two constructions are based on: 1. An asymmetric pairing group (G_1,\G_2), for which the explicit hardness of DDH holds in G_1 and the kernel Diffie-Hellman problem is hard in G_2. Explicit hardness states that the advantage of any PPT adversary is upper-bounded by an explicit bound \mu, which in our case is the inverse of an arbitrarily small superpolynomial function. The Kernel Diffie-Hellman problem is implied by the DDH assumption in the same group. 2. A pairing-free group G, in which the \mu explicit hardness of the DDH assumption holds relative to the inverse of an arbitrarily small superpolynomial function, along with the 2^{-\lambda/2}-OWKDM security of ElGamal with respect to efficient functions. Both ZAP constructions are encompassed in a paper published at TCC 2021.

Deep learning models have shown remarkable performance across various applications. However, their resilience to hardware faults, particularly the loss of neurons, is a critical aspect that needs to be addressed. In this paper, we propose a novel approach for improving the resilience of deep learning models. Through a carefully designed experiment, we evaluate the impact of neuron loss on these models in order to cultivate a deeper understanding of deep learning models' resilience and to identify methods for enhancing their reliability in the face of neuronal loss. This is done by investigating a new technique that focuses on pruning the most important neurons and then retraining the model to distribute the work among all the neurons. This technique along with dropout demonstrates great results that significantly enhanced resilience after neuron loss. This research bears significant importance, as deep learning models are increasingly applied across a wide range of domains, including computer vision, self-driving cars, and robotics. To ensure the dependability and stability of these systems, it is crucial to acknowledge their limitations and potential vulnerabilities, and to devise ways to strengthen their resilience. By examining the resilience of deep learning models concerning neuronal loss, we can gain valuable insights that will inform the design of more reliable artificial neural networks for real-world applications.

The limitation of DNNs to explain their black box behavior has led to a recent surge of explainability methods. However, there are growing concerns that these explainability methods are not robust and trustworthy. In this paper, we perform the first robustness analysis of Neuron Explanation methods under a unified pipeline and show that these explanations can be significantly corrupted by random noises and well-designed perturbations added to the probing dataset. We find that even adding small random noise with a standard deviation of 0.02 can already change the assigned concepts of up to 28% neurons in the deeper layers. Furthermore, we devise a novel corruption algorithm and show that our algorithm can manipulate the concept of more than 80% neurons by poisoning less than 10\% of probing data. This raises the concern of trusting Neuron Explanation methods in real-life safety and fairness critical applications.

Melanoma is an incredibly dangerous form of skin cancer. In 2023 alone, the American Cancer Society estimates that about 7,990 Americans will die from new melanoma cases. Image analysis tools that automate the diagnosis of melanoma will improve dermatologists' diagnostic accuracy and should provide a significantly better detection of melanoma. In this paper, we present an application of modern deep learning architectures using the images of skin lesions in combination with patient-level contextual information to detect the presence of melanoma. More specifically, we used the latest variant of the residual network, the ResNeSt model, to leverage its deep architecture and channel-wise attention mechanisms for melanoma detection. In parallel, we used a multi-layer perceptron model to process patient-level data, and fused its output with the extracted features from the ResNeSt model. To compare and contrast performances, we trained a similar model using a smaller, more generic, convolutional neural network in addition to the larger ResNeSt network. After properly tuning all associated hyperparameters, our primary model obtained a balanced accuracy of 0.853 and a ROC-AUC score of 0.93. The split-attention network was shown to improve the total ROC-AUC by approximately 9%, suggesting that our model was able to successfully leverage the ResNeSt design to extract the most important features and most relevant cross-feature interactions for the detection of melanoma.

The theoretical study of Message Passing Neural Networks (MPNN) focuses on increasing expressive power and overcoming over-squashing \& over-smoothing phenomena. The expressive power study of MPNN suggests that one needs to move from local computation to global modeling to gain expressive power in terms of the Weisfeiler-Lehman hierarchy. My dissertation centers around understanding the theoretical property of global GNN, its relationship to the local MPNN, and how to use GNN for coarse-graining. In particular, it consists of three parts:Convergence of Invariant Graph Network. In this part, we investigate the convergence of one powerful GNN, Invariant Graph Network (IGN) over graphs sampled from graphons. We first prove the stability of linear layers for general k-IGN (of order k) based on a novel interpretation of linear equivariant layers. Building upon this result, we prove the convergence of k-IGN under the model of \cite{ruiz2020graphon}, where we access the edge weight but the convergence error is measured for graphon inputs. Under the more natural (and more challenging) setting of \cite{keriven2020convergence} where one can only access 0-1 adjacency matrix sampled according to edge probability, we first show a negative result that the convergence of any IGN is not possible. We then obtain the convergence of a subset of IGNs, denoted as IGN-small, after the edge probability estimation. We show that IGN-small still contains functions rich enough to approximate spectral GNNs arbitrarily well. Lastly, we perform experiments on various graphon models to verify our statements.The Connection between MPNN and Graph Transformer. In this part, we study the connection between local GNN (MPNN) and global GNN (Graph Transformer). Graph Transformer (GT) recently has emerged as a new paradigm of graph learning algorithms, outperforming the previously popular Message Passing Neural Network (MPNN) on multiple benchmarks. Previous work \cite{kim2022pure} shows that with proper position embedding, GT can approximate MPNN arbitrarily well, implying that GT is at least as powerful as MPNN. In this paper, we study the inverse connection and show that MPNN with virtual node (VN), a commonly used heuristic with little theoretical understanding, is powerful enough to arbitrarily approximate the self-attention layer of GT.In particular, we first show that if we consider one type of linear transformer, the so-called Performer/Linear Transformer, then MPNN + VN with only $O(1)$ depth and $O(1)$ width can approximate a self-attention layer in Performer/Linear Transformer. Next, via a connection between MPNN + VN and DeepSets, we prove the MPNN + VN with $O(n^d)$ width and $O(1)$ depth can approximate the self-attention layer arbitrarily well, where d is the input feature dimension. Lastly, under some assumptions, we provide an explicit construction of MPNN + VN with $O(1)$ width and $O(n)$ depth approximating the self-attention layer in GT arbitrarily well. Graph coarsening with neural networks. In this part, we study the fundamental problem of reducing graph size/resolution while preserving its spectral property. We first propose a framework for measuring the quality of coarsening algorithm and show that depending on the goal, we need to carefully choose the Laplace operator on the coarse graph and associated projection/lift operators. Motivated by the observation that the current choice of edge weight for the coarse graph may be sub-optimal, we parametrize the weight assignment map with GNN and train it to improve the coarsening quality in an unsupervised way. Through extensive experiments on both synthetic and real networks, we demonstrate that our method significantly improves common graph coarsening methods under various metrics, reduction ratios, graph sizes, and graph types. It generalizes to graphs of larger size (25x of training graphs), is adaptive to different losses (differentiable and non-differentiable), and scales to much larger graphs than previous work.

At-home exercising strongly predicts physical therapy patient outcomes, underscoring the need for analyzing patient behaviors at-home via remote patient monitoring. Contemporary methods for remote patient monitoring rely on specialized sensors, i.e., Inertial Measurement Units, RGB-Depth cameras, motion capture systems, or stereo vision which are costly and not scalable to all physical therapy patients. Here, we observe a lack of literature using only a monocular RGB camera. In this paper, we demonstrate a skeletal feedback model for at-home exercises using only video acquired from a smartphone camera. We propose models for (i) Patient Performance Evaluation - which classifies the correctness of exercises, and (ii) Guidance - which identifies why the exercise went wrong so the patient can correct themselves. We use these models on our dataset of four common physical therapy exercises labeled by a physical therapist. Our results demonstrate the feasibility of using skeletal data from state-of-the-art 3D human pose estimation models for physical rehabilitation exercise evaluation and guidance. Thus, we enable remote patient monitoring and guidance from a single camera - making it highly cost-effective and scalable.

The interest in building blockchain Decentralized Applications (DApps) has been growing over the past few years. DApps are implemented as smart contracts which are programs that are maintained by a blockchain network. Building DApps, however, faces many challenges—most notably the performance and monetary overhead of writing to blockchain smart contracts. To overcome this challenge, many DApp developers have explored utilizing off-chain resources—nodes outside of the blockchain network—to offload part of the processing and storage.In this paper, we propose RollStore, a data indexing solution for hybrid onchain-offchain DApps. RollStore provides efficiency in terms of reduced cost and latency, as well as security in terms of tolerating byzantine (i.e., malicious) off-chain nodes. RollStore achieves this by: (1) a three-stage commitment strategy where each stage represents a point in a performance-security trade-off—i.e., the first stage is fast but less secure while the last stage is slower but more secure. (2) utilizing zero-knowledge (zk) proofs to enable the on-chain smart contract to verify off-chain operations with a small cost. (3) Combining Log-Structured Merge (LSM) trees and Merkle Mountain Range (MMR) trees to efficiently enable both access and verification of indexed data. We experimentally evaluate the cost and performance benefits of RollStore while comparing it with BlockchainDB and BigChainDB.

We introduce the problem of model-extraction attacks in cyber-physical systems in which an attacker attempts to estimate (or extract) the feedback controller of the system. Extracting (or estimating) the controller provides an unmatched edge to attackers since it allows them to predict the future control actions of the system and plan their attack accordingly. Hence, it is important to understand the ability of the attackers to perform such an attack. In this paper, we focus on the setting when a Deep Neural Network (DNN) controller is trained using Reinforcement Learning (RL) algorithms and is used to control a stochastic system. We play the role of the attacker that aims to estimate such an unknown DNN controller, and we propose a two-phase algorithm. In the first phase, also called the offline phase, the attacker uses side-channel information about the RL-reward function and the system dynamics to identify a set of candidate estimates of the unknown DNN. In the second phase, also called the online phase, the attacker observes the behavior of the unknown DNN and uses these observations to shortlist the set of final policy estimates. We provide theoretical analysis of the error between the unknown DNN and the estimated one. We also provide numerical results showing the effectiveness of the proposed algorithm.

In this paper we conduct transfer learning on the Faster R-CNN ResNet-50 FPN V2 object detection model from Benchmarking Detection Transfer Learning with Vision Transformers [12] to a new problem space: detecting what parts of a webpage can be clicked on. The model predicts the bounding boxes and class labels for what part of a website image can be clicked on, and what the result will be. We found that training the feature extraction layers of the model was detrimental to the learning (354% higher loss after 10 epochs), but that training the RPN layers was not detrimental. We additionally found that the most effective resizing for website imagery was to 224 x 224 pixels. Since the meta trainer we developed did not produce usable data, we experiment on using the data diversity approach in active learning to select the most representative images to train on. Compared to our control, the model trained on the active learning dataset scored 12.5% better on the validation data. However, with such a small dataset of 10 labeled images, the model was unable to generalize and did not predict any of the 15 objects on a labeled image outside the training set.

Generative artificial intelligence is strong at creative tasks. Image generation models like Stable Diffusion, Dall-E, and Midjourney generate images given a prompt. Language learning models like ChatGPT can generate text. Language learning models’ strengths are generating creative content like stories and poems. While current systems that combine the power of language learning models and image generation models to create storybooks exist, they lack image consistency - both in terms of style and visually depicting the text on the page.This paper introduces stability.gpt - A new modern system using the latest version of GPT (GPT-4) and Stable Diffusion to generate storybooks that have consistency in image style. Upon having some users create stories with stability.gpt and analyzing ten stories from a diverse set of genres, it becomes clear that while character consistency has improved and that the images support the story text, there is still room for improvement in generating consistent animals, objects, and styles.

This paper provides a pattern library of monetization methods that are common in games today, providing a resource for future designers and computational media, psychology, and business researchers to reference in their work. Such strategies as battle passes, loot boxes, premium currency, DLC, and subscriptions are discussed. The psychology, fairness, and narrative aspects of each strategy is examined in depth. The idea of games as a service is also examined as a framework for modern monetization models. A short history of games is also presented, to give background and context to the emergence of modern monetization models.

Trusted Execution Environment (TEE) is a blooming direction in the cloud industry. Aiming at protecting cloud user's data in runtime, TEE can enable a lot of new and foreseeable cloud use cases. While enclave-based TEEs such as Intel SGX suffer from a large effort of rewriting existing code, VM-based TEEs such as Intel Trust Domain Extension (TDX) and AMD Secure Encrypted Virtualization (SEV) attract more and more people's attention. Among those VM-based TEEs, AMD SEV is a security extension for the AMD Virtualization (AMD-V) architecture, which is AMD’s ambitious movement towards confidential cloud computing. SEV allows one physical server to efficiently run multiple guest virtual machines (VM) concurrently on encrypted memory with the goal of protecting the security of guest VMs even in the presence of a malicious hypervisor. SEV is also believed to be the first and the only commercial VM-based TEE that has already been adopted in Google Cloud and Microsoft Azure at the time of writing. However, the strong assumption of SEV causes uncertainty in its security guarantee. The lack of a systematic security study in this new assumption makes some unexploited vulnerabilities possible. Thus, it is very urgent to fully study SEV's design and help the community better understand SEV. In this dissertation, we systematically study the structure of AMD SEV's design, including angles from both hardware and software. By comprehensively exploring SEV's different components, we reveal how SEV's hardware and software work together to provide a trusted execution environment, and we also explore several unexploited vulnerabilities in SEV. Here, we briefly outline five categories of vulnerable designs in SEV and the corresponding security attacks. In Chapter 3, we exploit the unprotected I/O operations of SEV-enabled VMs and show that the malicious hypervisor can breach the confidentiality and the integrity of guest VMs with the help of these I/O operations.In Chapter 4, we explore the improper Address Space Identifier (ASID)-based memory isolation and access control. We show that the untrusted hypervisor has control over the VM's ASID without necessary hardware limits. We exploit this design and propose a series of attacks called CROSSLINE attacks. We show that this vulnerable design can be used to decrypt VM's encrypted memory or to momentarily execute arbitrary instructions of the victim VM. In Chapter 5, we provide the first exploration of TLB management in SEV. We first demystify how SEV extends the TLB implementation and show that the TLB management is no longer secure under SEV's threat model, which allows the hypervisor to poison TLB entries between two processes of a SEV VM. We then present TLB Poisoning Attacks, a class of attacks that break the integrity and confidentiality of the SEV VM by poisoning its TLB entries. In Chapter 6, we explore the context switch between the guest VM and the host. We show that during context switch, encrypting the virtual CPU's register stored in the VM Save Area is not enough, which allows the privileged adversary to infer the guest VM's execution states or recover certain plaintext. To demonstrate the severity of the vulnerability, we present the CIPHERLEAKS attack, which exploits the ciphertext side channel to steal private keys from the constant-time implementation of RSA and ECDSA in the latest OpenSSL library. In Chapter 7, we perform a comprehensive study of the ciphertext side channels that widely exist in almost all VM-based TEEs.Our work suggests that the deterministic encryption adopted by almost all commercial VM-based TEEs may potentially leak key-related secrets in most up-to-date mainstream cryptography libraries. The proposed generic ciphertext side-channel attack may exploit the ciphertext leakage from any memory page, including those pages used for kernel data structures, stacks, and heaps.This dissertation shows that the strong assumption of SEV indeed causes some unexploited vulnerabilities. We have reported our findings to the AMD SEV team and have received very positive feedback (several CVEs, two microcode patches, two cryptography library patches, three official security bulletins and an official white paper for software mitigation). We believe that our work can not only help improve the security of AMD SEV but also help the entire TEE community to build and develop more secure products for all cloud customers.

With the paradigm shift to cloud-based operations, reliable and secure access to and transfer of data between differing security domains has never been more essential. A Cross Domain Solution (CDS) is a guarded interface which serves to execute the secure access and/or transfer of data between isolated and/or differing security domains defined by an administrative security policy. Cross domain security requires trustworthiness at the confluence of the hardware and software components which implement a security policy. Security components must be relied upon to defend against widely encompassing threats -- consider insider threats and nation state threat actors which can be both onsite and offsite threat actors -- to information assurance. Current implementations of CDS systems use sub-optimal Trusted Computing Bases (TCB) without any formal verification proofs, confirming the gap between blind trust and trustworthiness. Moreover, most CDSs are exclusively operated by Department of Defense agencies and are not readily available to the commercial sectors, nor are they available for independent security verification. Still, more CDSs are only usable in physically isolated environments such as Sensitive Compartmented Information Facilities and are inconsistent with the paradigm shift to cloud environments. Our purpose is to address the question of how trustworthiness can be implemented in a remotely deployable CDS that also supports availability and accessibility to all sectors. In this paper, we present a novel CDS system architecture which is the first to use a formally verified TCB. Additionally, our CDS model is the first of its kind to utilize a computation-isolation approach which allows our CDS to be remotely deployable for use in cloud-based solutions.

Due to its potential improvement on network throughput, network coding hasattracted considerable research interests. Random Linear Network Coding (RLNC),as a branch of research on network coding, is proposed as a cipher to protect theconfidentiality of packets possibly on the future Internet due to its features suchas packet mixing in a decentralized approach. In this paper, we propose attacksto compromise RLNC as a cipher for confidentiality protection. The attacks arebased on the blind source separation (BSS) technique, a statistical signal processingtechnique designed to recover original signals based on mixtures of original signals.We also design the scaling step to filter out artifacts generated by BSS algorithms andthe cross-checking steps to increase confidence on packets recovered by the proposedattacks. Our extensive experiments on packets collected from the Internet and acampus network show that the attacks can successfully recover about 1% of originalpackets.

The emergence and evolution of ideas is one of the most important processes in human society, and has been a topic of great interest for philosophers and historians. Psychologists have also attempted to develop models of how new ideas arise from the recombination of existing ones, and have proposed to model this process as being similar to biological evolution. However, studying the evolution of ideas has been limited by the difficulty in obtaining systematic data. The recent exponential growth in electronic data promises a solution, but several impediments remain, including a systematic process for extracting ideas and methods for analyzing their dynamics over time. While the general problem of identifying ideas in texts is extremely complex, one possible approach is to look at how meaning is distributed in documents, and to study the evolution of this structure across documents and over time. The research in this dissertation develops a framework for doing this in large, longitudinal corpora of documents. This system, calledthe Framework for the Analysis of Semantic Structure Evolution in Text (FASSET),exploits the statistics of changing word usage within the corpus in combination with machine learning techniques, including topic analysis, semantic embedding, adaptive clustering, and dimensionality reduction for visualization. It represents an integrated model for extracting semantic structure within single documents, within text corpora with a single time-stamp, and across a longitudinally extensive corpus. It includes new methods for text segmentation, longitudinal topic identification, and longitudinal semantic clustering. The goal is to provide a system for exploring a simplified "systems biology" of ideas through which the evolution of ideas can be studied at various levels.The FASSET system is applied to two large longitudinal corpora: Speeches in the U.S. Congress over a period of 36 consecutive years, and papers presented at the International Joint Conference on Neural Networks (IJCNN) over 18 consecutive years. The results show what issues and ideas have animated the areas of American politics and neural networks over time, how various topics have evolved in that time, and how new topics have emerged from them.

Kirigami metamaterials are engineered materials inspired by the ancient art of paper cutting. To produce them, partial cuts are placed in a base material so that stretching the material causes it to deploy, or change size and shape. Other exotic mechanical properties can emerge as well. These materials are applicable for many engineering settings where controllable structural change is desired. Existing research has focused on how kirigami patterns based on simple square and triangle grids behave when cut into different materials or used in different settings. In this thesis, we seek to understand the space of kirigami patterns beyond those based on grids. How do we find these patterns, what new features could they offer, and what useful information might lie in a pattern’s underlying structure? First, we find a set of kirigami patterns representing the 17 wallpaper groups, which are symmetry groups characterizing the space of 2D periodic patterns. By studying the graphs representing these initial kirigami patterns’ underlying connectivity, we develop a pattern augmentation method that can 1) produce new kirigami patterns, 2) help us deploy tilings that were previously difficult or impossible to deploy, and 3) produce large and controllable size change in kirigami patterns. We also study how the symmetries of a kirigami pattern change through deployment. In the second half of our work, we develop kirigami patterns based on aperiodic quasicrystal tilings. Since these tilings never repeat, they require developing more general kirigami design principles based on underlying structure. We introduce three such design methods and develop a computer simulation for verifying their effectiveness. We analyze geometric, mechanical, and topological properties of our methods and show that each method has different properties suitable for different applications. Finally, we show that one of our deployment methods preserves aspects of the unique long-range order of quasicrystal tilings.

In this paper, we provide an overview of the key concepts and data structures that underly Fxfs, one of Google’s newest open-source filesystems. Compared to traditional filesystems such as ext4, Fxfs has an unconventional design as it relies heavily on log-structured merge (LSM) trees to store its metadata. Every filesystem needs an allocator to determine where and how to store files on a device. The existing Fxfs allocator takes a partial fit approach and returns the first free gap that it finds on the device. There is a potential opportunity to improve the performance of the filesystem by designing an allocator that reduces fragmentation. The hypothesis is that with less fragmentation, an allocator will issue fewer disk IOs, which could lead to performance benefits. To test this hypothesis, we implement four new allocators and run them all on a sqlite benchmark. The results suggest that the time spent allocating is a significant driver of performance and should be considered as one of the key factors while designing a strong allocator.

Differential privacy (DP) has a strong theoretical foundation, but the path to implementing DP functions on computers is fraught with challenges. A common approach for creating a DP version of some function is to make a two-part algorithm: the algorithm first computes the true result of the function on a dataset, and the algorithm then draws noise from some probability distribution -- with the scale of this distribution based on the behavior of the function which computes the true result -- to perturb this true result and return an answer that both preserves the privacy of individuals in the dataset and provides utility for data analysts. Previous research has investigated the effects of finite-precision arithmetic on the ``noise-generation step'' of a DP algorithm. These efforts have revealed that finite-precision analogs of noise functions can behave far differently from their continuous counterparts, and that these differences can produce exploitable privacy violations. However, there has been little investigation into the effect of finite-precision arithmetic on the ``true answer step'' of a DP algorithm. In this paper, we describe how the true answer step can also behave far differently than expected, and how these differences in behavior can result in exploitable privacy violations. We demonstrate how a malicious data analyst can take advantage of these differences in behavior and use the outputs of supposedly DP functions to perform membership-inference attacks and reconstruction attacks on datasets. (These attacks affect all of the libraries of general-purpose DP functions that we investigated, including Google's DP library, IBM's diffprivlib, and Microsoft's SmartNoise.) We also describe how the functions in these libraries can be modified to offer privacy, and we prove the privacy of these modified functions.

Data scientists work with large-scale sensitive data, which inevitably leads to privacy risks. Differential Privacy (DP) is a mathematical definition of privacy that aims to mitigate privacy risks inherent in data analysis and machine learning. OpenDP, an open-source software DP tool, allows the government, industry, and academic institutions to share sensitive data to researchers or the public while preserving privacy. An active research question in DP literature is how to bound the total privacy loss of a sequence of DP computations. Most known DP theorems and the OpenDP library require the privacy parameters of each computation to be fixed in advance. However, this prevents the design of privacy preserving machine learning algorithms that change the privacy budget on the fly. For the adaptive parameter setting, privacy filters and odometers are two objects designed to model the total privacy loss. In this paper, we extend the programming framework for OpenDP library to handle DP composition under adaptive privacy budgets through Renyi Differential Privacy (RDP). To do so, we construct Renyi filter and odometer and prove its privacy guarantees by generalizing RDP Adaptive Composition. To generalize our Odometer results, we implement a constructor that converts any Odometer to a Filter. Our results allow for the real world DP deployment of ML algorithms and interactive query interfaces that adaptively update the privacy budget.

COVID-19 pushed millions of workers from the office to Zoom. With remote work becoming the new normal, firms and employees alike continue to wonder about the benefits or drawbacks to remote work on overall productivity. This paper presents a model for a firm of information workers, like that of Microsoft or Apple, and an understanding for how the productivity of such a firm reacts to a shift to remote work. Using real world effects from Microsoft’s own company-wide shift to remote work, we model productivity of a spectrum of firms pre- and post-remote work along with various formal and informal firm structures. We find that across firm types, productivity, as measured by a time-based metric of job-completion for an innovation task and a random task, significantly decreased in a post-remote work world. However, we find that instituting a simple manager network can help ensure a firm is robust to future shifts to remote work. Likewise, encouraging employees to regularly meet with individuals of other worker types and creating large employee gatherings are similarly robust methods of insulating a firm from remote work shocks. The creation and understanding of this model of a firm and its interventions opens up new opportunities for understanding the interplay between intra-/inter- group collaboration, formal/informal firm network dynamics, and helping firms and employees alike navigate a new era of remote work.

This dissertation explores how computer science became a discipline in the United States and how this process of discipline formation and institutionalization shaped the content and culture of computer science. Like other fields of knowledge, computer science in the United States emerged out of established disciplines; however, unlike many other scientific or engineering fields, computer science developed simultaneously out of multiple other disciplines and multiple different interdisciplinary arrangements. This dissertation explores how each of these disciplines in specific university contexts contributed to the construction of the character of computer science and the definition of the identity of the computer scientist and how different interdisciplinary relationships provided opportunities for the development of distinct visions of computer science. Drawing on archival research from university and department records as well as the papers of university administrators, computer scientists, mathematicians, electrical engineers, and others, this work reveals how the processes of discipline formation shaped everything from the intellectual foundation and internal culture of computer science to the modern American university’s institutional structure and approach to interdisciplinary knowledge-making.