Today's data-driven applications, such as big data analytics, neural networks, and machine learning, require huge memory and computing resources. The latency and energy incurred for data movement between processing unit and memory has become a significant limitation (known as the "memory wall") as traditional techniques such as caching are no longer effective for the data-intensive applications that dominate modern computing. Solving the "memory wall" problem requires future computer technology to directly integrate memory and transistor devices with high density vertical interconnect accesses (vias) to provide parallel, high bandwidth memory access. However, the high temperature of silicon processing is not compatible with this 3D monolithic integration. A desired low temperature integration can be achieved by using two-dimensional (2D) materials which have intrinsically layered and flat atomic structures, because devices made of 2D materials can be fabricated at lower temperatures. This thesis introduces three exploratory studies on advancing 2D materials for developing 3D monolithic integrated systems. First, I focused on improving transistor performance and studied high-mobility 2D material black phosphorous (BP) transistors using various metals as the contact metal. This work achieved unipolar n-type BP transistors by using ultra-low work function metals, demonstrating record high n-type current. Furthermore, the study revealed the physical mechanisms of controlling doping and de-pinning effects for n- and p- type BP transistors. Beyond transistor studies, I then demonstrated the first 3D sequential monolithic integrated two levels of 1-transistor-1-resistor (1T1R) memory cells. The cell is fabricated entirely using 2D materials: hexagonal-Boron Nitride (hBN) serves as the resistive switching memory cell and monolayer Molybdenum Disulfide (MoS2) serves as the channel material for the transistor selector. In order to achieve a high-density memory array, a two-terminal selector with a high nonlinear current-voltage characteristic is necessary. I developed a new two-terminal selector utilizing a 2D material heterostructure with an H-shape energy barrier. This new device design in theory should have high on-state current density and virtually unlimited endurance due to its quantum tunneling mechanism. Our results characterized the first out-of-plane current through an ultra-thin (3 monolayer) heterojunction and provide the critical foundation for a high endurance selector using a 2D heterojunction. Due to the transferable feature of 2D materials, these processes can be easily adopted for 3D monolithic integrated transistors and memories in multiple logic and memory layers connected vertically by fine-grained nanoscale vias, thereby overcoming the "memory wall.".

Human cognition is fundamentally flexible — we can adapt to novel tasks rapidly. We can sometimes adapt to a novel task without any direct experience on that task, based on its relationship to previous tasks. By contrast, while deep-learning models can achieve superhuman performance on many tasks, they are often unable to adapt to even slight task alterations. This ostensible inflexibility has led to criticism of deep learning models by cognitive scientists. I begin this dissertation by reviewing the literature on cognitive flexibility, and recent advances in building more flexible artificial intelligence systems. I provide a synthesis of these literatures, and outline the challenges that I believe remain. In particular, I focus on the ability to adapt to new tasks zero-shot — that is, without any data — based on their relationship to prior tasks. To address this challenge, I propose a general computational framework for adaptation to novel tasks based on their relationship to prior tasks. The framework is based on meta-mappings, higher-order tasks that transform basic tasks. I propose a parsimonious implementation of this framework in the form of homoiconic meta-mapping architectures. I demonstrate this framework across a wide variety of tasks and computational paradigms, ranging from regression to image classification and reinforcement learning. I compare to both human adaptability, and language-based approaches to zero-shot task performance. I show that meta-mapping is quite succesful, often achieveing 80-90% performance on a novel task, even when the new task directly contradicts prior experience. I further show that using this adaptation as a starting point can dramatically accelerate later learning on a task, and reduce the errors made on the way to mastery by nearly an order of magnitude. Thus, I suggest that meta-mapping can provide a computational basis for adapting to new tasks, and a starting point for efficient learning. This dissertation therefore provides a framework for building better cognitive models and more flexible artificial intelligence systems. In the final chapter, I review the broader contributions of this work to an ongoing discussion about the computational principles necessary for intelligence, and highlight possible future directions ranging from understanding mathematical cognition to neuroscience.

Magnetic resonance imaging (MRI) is an important medical imaging modality for imaging soft tissue. A fundamental limitation of MRI is that there are tradeoffs between scan time, image resolution, and image signal-to-noise ratio (SNR). Typically, image resolution and SNR are not sacrificed because the images need to be diagnostically useful. Therefore, scan time is typically increased. For magnetic resonance angiography (MRA), this problem is exacerbated. Blood vessels are very small and, because of the ballistics of the heart pumping, the arteries are constantly in motion. Therefore, high resolution is required to resolve the vessels and the images need high SNR such that the vessels are distinguishable from noise. Additionally, while most clinical MRI acquires 2D slices, MRA typically requires 3D imaging because blood vessels typically do not lie in a single plane, which further extends scan time. To accelerate 3D MRA, we approached the problem from several directions. One method was to reduce the signal from all of the other tissue that we did not wish to image. We proposed a combined T2-preparation, outer volume suppression (OVS), fat saturation pulse to improve 3D coronary artery imaging. The pulse sequence consisted of the following pulse sequence: a 90°_-60 180°_{60} composite nonselective tip-down pulse, two 180°_Y hard pulses for refocusing, and a -90° spectral-spatial sinc tip-up pulse. The pulse sequence was played twice, first selective in the x-axis and then selective in the y-axis for a 2D OVS. Compared to another published T2-preparation OVS pulse sequence, the proposed pulse had superior image edge profile acutance values for the right (P< 0.05) and left (P< 0.05) coronary arteries, suggesting superior vessel sharpness. The proposed sequence also had superior SNR (P< 0.05) and passband-to-stopband ratio (P< 0.05). Reader scores and reader preference indicated superior image quality of the proposed sequence for both the right (P< 0.05) and left (P< 0.05) coronary arteries. Therefore, the proposed sequence had superior image quality and suppression, potentially enabling scan acceleration with reduced field of view imaging. Another technique to accelerate MRA was to enable more efficient acquisition trajectories. Non-Cartesian trajectories have the potential to be highly efficient but their efficiency is limited in practice by off-resonance. A residual convolutional neural network was trained to correct off-resonance artifacts in pediatric MRA exams (Off-ResNet). Training data was acquired from exams with a short readout scan (1.18 ms ± 0.38) and a long readout scan (3.35 ms ± 0.74) at 3 T. Short readout scans with longer scan times but negligible off-resonance blurring were used as reference images and augmented with additional off-resonance for supervised training examples. Long readout scans with greater off-resonance artifacts but shorter scan time were corrected by autofocus and Off-ResNet and compared with short readout scans by normalized root-mean-square error (NRMSE), structural similarity index (SSIM), and peak SNR (PSNR). Scans were also compared by scoring on eight anatomical features by two radiologists. Off-ResNet had superior NRMSE, SSIM, and PSNR compared to uncorrected images across ±1 kHz off-resonance (P< 0.01). Off-ResNet also had superior NRMSE over -677 Hz to +1 kHz and superior SSIM and PSNR over ±1 kHz compared with autofocus (P< 0.01). Radiologic scoring demonstrated that long readout scans corrected with Off-ResNet were noninferior to short readout scans (P< 0.05). The long readout scans were 59.3% shorter than the short readout scans, and Off-ResNet was able to correct the long readout scans to a noninferior quality compared to the diagnostically standard short readout scans. One limitation of Off-ResNet was that it used zero-order field maps to generate training data. While zero-order field maps are not realistic, it is also expensive to acquire large datasets with field maps. A generative adversarial network (GAN) was trained to generate realistic but not quantitatively accurate field maps for a given anatomical input image. The input to the generator network was an anatomical image from a single echo of a multi-echo T2*-IDEAL scan and the real examples of the corresponding field maps were also from the same set of T2*-IDEAL scans. We also proposed using a discriminator with Off-ResNet to improve off-resonance correction (Off-ResGAN). To evaluate the contributions of using a dataset with generated field maps and a GAN for off-resonance correction respectively, we trained four networks: 1. Off-ResNet and a dataset created with zero-order field maps, 2. Off-ResNet and a dataset created with generated field maps, 3. Off-ResGAN and a dataset created with zero-order field maps, and 4. Off-ResGAN and a dataset created with generated field maps. The four networks were evaluated on validation datasets created with zero-order field maps and generated field maps. The Off-ResGAN networks had superior NRMSE, SSIM, and PSNR on both datasets. With respect to datasets, each network performed best on the validation dataset that matched its training dataset. However, the networks trained with zero-order field maps tended to have worse performance on the generated field map validation dataset relative to the the performance of networks trained with generated field maps on the zero-order field map validation dataset. These initial results suggested that Off-ResGAN has superior image quality, and the benefit of using a dataset with generated field maps is increased robustness with respect to realistic off-resonant images. Undersampling is an important technique to accelerate scans. Undersampling reconstruction can involve a combination of parallel imaging and compressed sensing. An open question however is an optimal undersampling trajectory as measured by an image quality metric and with respect to a reconstruction technique. Moreover, as the scan proceeds, more information is collected and therefore increasingly better decisions can be made about the optimal undersampling trajectory. We proposed an approach using deep reinforcement learning to train an agent to learn from previous scans and combine the previously learned statistics with the new incoming data to determine the trajectory of the next readout. To frame the reinforcement learning problem, the environment was the reconstruction technique, the reward was the image metric, and the agent was a deep Q-learning convolutional neural network. Initial results with an L2 image quality metric and compressed sensing and deep learning reconstruction techniques indicated that the agent was capable of learning optimal trajectories. L2 was a special metric where we could analytically calculate the optimal trajectory. This suggested that the agent could learn optimal trajectories for other image metrics and reconstruction techniques where computing the optimal trajectory was intractable. In conclusion, the proposed techniques have demonstrated several strategies for accelerating 3D MRA. Through a combination of them, 3D MRA may become more feasible and accessible as a clinical tool

In the last decade, online learning platforms have dramatically increased access to educational materials, and replaced and supplemented traditional co-located instruction. Throughout this shift, instructors have struggled with how best to create and deploy materials to online platforms and understand their effectiveness when scaled across hundreds or thousands of learners. To address these complementary problems, this thesis introduces new systems and methods for both learning analytics and adaptive instruction. In OARS, we demonstrate a real-time learning analytics system deployed across more than ten online courses with tens of thousands of learners. We then report on the value of learning analytics from the perspective of course instructors. Learning from these perspectives, we next introduce a system for adaptively scheduling educational activities through reinforcement learning (RL). Without any skill labels, this model learns how to assign educational activities in a way that maximizes learning gains while minimizing redundant work. Finally, in a randomized controlled experiment, we show that our RL scheduling algorithm helped to improve the educational experience for online learners by reducing the number of learning activities required to produce the same learning gains

The first published implementation of a binding assay was an immunoassay for insulin detection in 1960. Over the subsequent 60 years, analytical technologies based on binding assays have evolved substantially. Molecular detection platforms based on enzyme-linked immunosorbent assays (ELISAs)—as well as newer commercial technologies, such as Luminex and NanoString—are core components of today's diagnostic and research armamentaria. There are two main challenges in the development of binding assays. First, binding alone does not generate an observable signal. The target and/or affinity reagent must be directly coupled to a moiety that generates a signal upon binding. Myriad immunosensors have been developed that couple the interaction between an antibody and its antigen to an observable output based on an electronic, optical, or mechanical signal. The second challenge is that binding assays are governed by the principles of chemical equilibrium. Binding assays generate signal through concentration-dependent shifts in equilibrium that result from the interaction of the target molecule with an affinity reagent. Convention affixes the range of measurable target concentrations to an 81-fold dynamic range centered around the dissociation constant of an affinity reagent to its target. In this dissertation, I introduce three major advances to the development of binding assays: (i) I demonstrate that high-throughput DNA sequencing can be used as a readout mechanism that is in many ways superior to optical and electrical readouts, (ii) I describe novel methodologies of modulating detection ranges without changing the actual physicochemical interactions between target and affinity reagent, and (iii) I challenge the conventional wisdom regarding binding assays and introduce a new conceptual framework for developing modern analytical technologies. Together, these studies provide new insights into the use of binding assays for molecular quantification and represent a rich resource for understanding the design of future assays

Nature has bestowed upon hummingbirds and other nectarivorous animals a singular ability to sip energy rich nectar by hovering in front of flowers. However, these flowers are not static: they have different orientations and move in unpredictable ways due to environmental wind as well as self-induced wind from the hovering animal. Hovering is unstable and extraordinarily energetically costly, so these animals must be able to track the moving flowers effectively. To study this phenomenon, we built a robotic flower that moves laterally and longitudinally to investigate how hummingbirds track moving flowers and how this compares to other animals, as it has been shown that hawkmoths can effectively track laterally-moving flowers in natural wind. Sound plays a key role in animal survival, as it is produced naturally during locomotion. Some animals maximize these sounds for communication, while others minimize them for predator avoidance. We performed nearfield acoustic holography measurements on hovering and maneuvering hummingbirds tracking moving flowers, for which we found hummingbirds modulate their wingbeat to track moving objects. We created a first principles acoustic model to further consider how hummingbirds hum and generate sound, which illustrated their hum can be explained by the sound associated with moving aerodynamic forces (lift and drag) acting on each wing. This model is expanded to include other flapping wing animals, ranging from fruit flies to parrotlets, allowing us to investigate effects over a wide range of wingspans, masses, flapping frequencies, and flapping styles. We found an important link between the weight support profile and radiated power: higher harmonic frequency content in the weight support profile means the animals radiates more acoustic power, which is important for some insects like flies which communicate with their wing noise and have a rich amount of harmonic content in their wingbeat. Over the orders of magnitude considered we found that the radiated acoustic power scaled most strongly with weight, then wingspan, and finally wingbeat frequency. While the acoustics can inform us about the forces indirectly through pressure, we would like to directly measure the in vivo lift and drag forces, especially since acoustic methods are unable to measure DC pressure offsets. To accomplish this feat, we developed a new instrument: the 3D Aerodynamic Force Platform (AFP). The flapping animal generates a pressure field which travels to the boundaries of the rectangular flight arena at the speed of sound. Each of the six instrumented plates mechanically integrates the pressure distribution, which is measured by three force sensors on each plate. By using an AFP, we can measure the forces associated with maneuvering while tracking a moving flower. Wing kinematics can be captured in conjunction by using an array of 3D-calibrated high-speed cameras. While flower orientation is important ecologically for the coevolution of flowers and hummingbirds, it is unclear how the forces are associated with that change. While horizontal flowers offer a single entry point for the hummingbird to feed, and through evolution can optimize the location of their stamen to increase the chances of a successful pollination, vertically oriented flowers lack this degree of freedom. These flowers instead have a radially symmetric stamen orientation. Flowers oriented vertically upwards have been shown to be more attractive to nectarivores, while the majority of downwards facing (pendant) flowers are pollinated solely by hummingbirds. We compared the power requirements for hummingbird feedings from horizontal flowers, upwards flowers, and pendant flowers. Hummingbirds incur a 14% increased aerodynamic power requirement for feeding from pendant flowers and a 25% increase for feeding from upwards flowers

Emotions play a central role in the human's ability to understand the world as well as in affecting consumers' perception of products, which further determine purchase intention and consumers' decisions to trust a product. Creating a specific emotional connection between consumers and products can enrich user experience, enhance product adoption, and increase user-to-product trust. Therefore, it is important for engineers and designers to understand how consumers perceive products regarding emotional connections and whether or not emotive stimuli affect consumers' perception and trust level of products. Insights in understanding emotional connections and effects of emotions on trust will help designers to effectively bond consumers and products and will also enable designers in calibrating user-to-product trust to a healthy level. To study the emotional connections that relate to use-to-product trust, the work presented in this dissertation employed a collage tool and the psychological priming technique in two different ways: first, the collage tool was used to examine how consumers evaluate products concerning emotional connections; second, the priming technique was applied to evoke designated emotions and stimulate a desirable design mindset. Currently, industrial designers rely on artistic intuition to evoke specific emotions in product aesthetics. A more systematic method is needed to measure emotional responses and to understand the association between product features and emotions. Study 1 in Chapter 2 proposed a two-axis collage tool to measure emotional connections in a visual and interactive way. Using the collage tool, the study participants evaluated a variety of wearable products and determined whether or not they perceived the wearables as comfortable, delightful or useful. In addition to determining purchase intention, emotions can potentially influence users' decisions to trust by affecting their mental state and perception of products. Derived from theories of interpersonal trust, trusting products involves both cognitive and affective processes—users assess a product's ability using analytic thought processes, and also weigh affective factors such as emotional investments, concerns for others and the beliefs of reciprocity of sentiment. Study 2 in Chapter 3 investigates the affective process of forming trust in interactive products, so that designers can manipulate users' trust to prevent misuse and disuse. The study investigated the affective process of trust formation of the Amazon Echo, an autonomous voice-activated assistant, by priming study participants with emotive images prior to interacting with the Echo. As priming successfully changed user-to-product trust by influencing users' mental state, the priming technique can also be applied to influence designers' mindsets. Study 3 in Chapter 4 applied priming to alter designers' mindsets prior to conceptual design exercises and investigated whether or not priming method improved ideation outcomes in terms of relevance to the sustainability pillars and ideas' intrinsic quality, as judged by experts, and also whether or not it caused designers to decrease the favoring of one's own ideas compared to the ideas of others during idea evaluation. This dissertation demonstrates the effect of emotions on users' perception of products and also the effect of manipulating mental state or mindset on users' decision and designers' capability. The results underscore the importance of understanding the mental model and the subconscious mindset of both users and designers in design processes. Insights from this research will help build academic theory and design guidance about how to ensure future technology to be socially desirable, collaborative and sustainable

This dissertation analyzes the literary response to the 2008 financial crisis in Spain. I argue that literature serves as a powerful tool to capture what is all too often left behind in discussions of the crisis. The texts analyzed in this dissertation show that the 2008 crisis was a symptom of deeper issues: the legacy of the Spanish transition and its links to corruption and crony capitalism; the precarity that defines the modern labor environment; and the tendency of official discourses to divert attention from emerging crises. I draw on the writings of Karl Marx and Adam Smith to discuss the inner workings of capitalism and its propensity for crisis. In conclusion, I put forth that by looking to the past, the literature of the 2008 financial crisis in Spain teaches us important lessons for the future

As the twin movements of open science and open source bring an ever greater share of the scientific process into the digital realm, new opportunities arise for the meta-scientific study of science itself, including of data science and statistics. Future science will likely see machines play an active role in processing, organizing, and perhaps even creating scientific knowledge. To make this possible, large engineering efforts must be undertaken to transform scientific artifacts into useful computational resources, and conceptual advances must be made in the organization of scientific theories, models, experiments, and data. This dissertation takes steps toward digitizing and systematizing two major artifacts of data science, statistical models and data analyses. Using tools from algebra, particularly categorical logic, a precise analogy is drawn between models in statistics and logic, enabling statistical models to be seen as models of theories, in the logical sense. Statistical theories, being algebraic structures, are amenable to machine representation and are equipped with morphisms that formalize the relations between different statistical methods. Turning from mathematics to engineering, a software system for creating machine representations of data analyses, in the form of Python or R programs, is designed and implemented. The representations aim to capture the semantics of data analyses, independent of the programming language and libraries in which they are implemented

This project tells a history of allegory from late modernism into the present. It presents case studies of four sites of allegorical thought and practice in late twentieth-century American cultural production, with a focus on post-WWII art, theory, literature, and material culture: Aldous Huxley's psychedelic writings in dialogue with the work of Color Field painter Morris Louis and his critics; the relationship between materiality and hermeneutics in Ken Kesey's communal art project, Furthur; the literary theory of Paul de Man, where allegory becomes conflated with irony; and allegorical structure and performativity in David Foster Wallace's Infinite Jest. Understanding allegory as a structure of thought—a transmedial structure based on an urge for meaning-making inherent in the human mind that is modified and adapted in response to evolving historical contexts, circumstances, and pressures—it examines these sites of allegorical transformation to ask how allegory was mobilized, theorized, or used by historical actors working within the limits of their contextual conditions of possibility. In doing so, it asks how recent changes in allegorical form, structure, and function come to shape the mode's contemporary appearance, arguing that allegory's structure undergoes a progressive flattening of levels as its aesthetic transcendence is reimagined as aesthetic immanence

In this thesis, we study two linearized approximations to neural networks: Random Feature model (RF) and Neural Tangent Model (NT). For two-layer neural networks (NN), we prove the existence of a performance gap between NN and these approximations under two simple data distribution models. We show that this gap in performance stems from the fact that, unlike NT and RF, NN learns efficient representations of the target function. In the second part of this thesis, we study the approximation power of NT and RF in high dimensions. We show that when the features are uniformly distributed on the sphere, these models can only fit surprisingly simple polynomial functions to the data. In the end, we examine the generalization behavior of these approximations in high dimensions. We show that in order to learn anything beyond low-degree polynomial functions, these approximate networks require extremely large number of training data points. Our results suggest that, while these approximations might perform well in various applications, they do not sufficiently capture the full power of neural networks