150813 research outputs found
Sort by
Do We Learn From Each Other: Understanding the Human-AI Co-Learning Process Embedded in Human-AI Collaboration
No full textBeyond collaborating in the AI-supported decision-making setting to achieve complementary performance, human and AI should learn from each other and internalize knowledge from their collaboration. This can enhance their individual performance when working independently after their collaboration. However, this expected dual-pathway co-learning process, including both “human learns from AI” and “AI learns from human”, does not occur spontaneously. Human-AI collaboration designs could have inconsistent and intertwined influences on the co-learning process. Based on the learning cycle theory, this study conducted three online, two-stage, and between-subject behavioral experiments to reveal how human and AI learn from each other. By developing a context where human and AI have comparable and moderate performance on emotion classification tasks, our study provides the first empirical evidence of an effective human-AI co-learning process within human-AI collaboration. However, the AI feedback and collaborative workflow design can lead to unequal and potentially negative impacts on both pathways of the co-learning process in groups with varying levels of cognitive reflection capability. These findings highlight three design principles to facilitate the co-learning process embedded in human-AI collaboration rather than naively deploying a complex AI system
Factorization and Compositional Generalization in Diffusion Models
No full textOne of the defining features of human intelligence is compositionality—the ability to generate an infinite array of complex ideas from a limited set of components. This capacity allows for the creation of novel and intricate combinations of arbitrary concepts, enabling potentially infinite expressive power from finite learning experiences. A likely prerequisite for the emergence of compositionality is the development of factorized representations of distinct features of variation in the world. However, the precise mechanisms behind the formation of these factorized representations in the human brain, and their connection to compositionality, remain unclear. Diffusion models are capable of generating photorealistic images that combine elements not co-occurring in the training set, demonstrating their ability to compositionally generalize. Yet, the underlying mechanisms of such compositionality and its acquisition through learning are still not well understood. Additionally, the relationship between forming factorized representations of distinct features and a model’s capacity for compositional generalization is not fully elucidated. In this thesis, we explore a simplified setting to investigate whether diffusion models can learn semantically meaningful and fully factorized representations of composable features. We conduct extensive controlled experiments on conditional diffusion models trained to generate various forms of 2D Gaussian data. Through preliminary investigations, we identify three distinct learning phases in the model, revealing that while overall learning rates depend on dataset density, the rates for independent generative factors do not. Moreover, our findings show that models can represent continuous features of variation with semi-continuous, factorized manifolds, resulting in superior compositionality but limited interpolation over unseen values. Based on our investigations, we propose a more data-efficient training scheme for diffusion models and suggest potential future architectures for more robust and efficient generative models.S.M
Cloudy skies: assessing public understanding of global warming
No full textSurveys show that most Americans believe global warming is real. But many advocate delaying action until there is more evidence that warming is harmful. The stock and flow structure of the climate, however, means “wait and see” policies guarantee further warming. Atmospheric CO2 concentration is now higher than any time in the last 420,000 years, and growing faster than any time in the past 20,000 years. The high concentration of CO2 and other greenhouse gases (GHGs) generates significant radiative forcing that contributes to warming. To reduce radiative forcing and the human contribution to warming, GHG concentrations must fall. To reduce GHG concentrations, emissions must fall below the rate at which GHGs are removed from the atmosphere. Anthropogenic CO2 emissions are now roughly double the removal rate, and the removal rate is projected to fall as natural carbon sinks saturate. Emissions must therefore fall by more than half even to stabilize CO2 at present record levels. Such reductions greatly exceed the Kyoto targets, while the Bush administration's Clear Skies Initiative calls for continued emissions growth. Does the public understand these physical facts? We report experiments assessing people's intuitive understanding of climate change. We presented highly educated graduate students with descriptions of greenhouse warming drawn from the IPCC's nontechnical reports. Subjects were then asked to identify the likely response to various scenarios for CO2 emissions or concentrations. The tasks require no mathematics, only an understanding of stocks and flows and basic facts about climate change. Overall performance was poor. Subjects often select trajectories that violate conservation of matter. Many believe temperature responds immediately to changes in CO2 emissions or concentrations. Still more believe that stabilizing emissions near current rates would stabilize the climate, when in fact emissions would continue to exceed removal, increasing GHG concentrations and radiative forcing. Such beliefs support “wait and see” policies, but violate basic laws of physics. We discuss implications for education and public policy
Telehealth and Virtual Reality Technologies in Chronic Pain Management: A Narrative Review
No full textPurpose of Review This review provides medical practitioners with an overview of the present and emergent roles of telehealth and associated virtual reality (VR) applications in chronic pain (CP) management, particularly in the post-COVID-19 healthcare landscape. Recent Findings Accumulated evidence points to the efficacy of now well-established telehealth modalities, such as videoconferencing, short messaging service (SMS), and mobile health (mHealth) applications in complementing remote CP care. More recently, and although still in early phases of clinical implementation, a wide range of VR-based interventions have demonstrated potential for improving the asynchronous remote management of CP. Additionally, VR-associated technologies at the leading edge of science and engineering, such as VR-assisted biofeedback, haptic technology, high-definition three-dimensional (HD3D) conferencing, VR-enabled interactions in a Metaverse, and the use of wearable monitoring devices, herald a new era for remote, synchronous patient-physician interactions. These advancements hold the potential to facilitate remote physical examinations, personalized remote care, and innovative interventions such as ultra-realistic biofeedback. Despite the promise of VR-associated technologies, several limitations remain, including the paucity of robust long-term effectiveness data, heterogeneity of reported pain-related outcomes, challenges with scalability and insurance coverage, and demographic-specific barriers to patient acceptability. Future research efforts should be directed toward mitigating these limitations to facilitate the integration of telehealth-associated VR into the conventional management of CP. Summary Despite ongoing barriers to widespread adoption, recent evidence suggests that VR-based interventions hold an increasing potential to complement and enhance the remote delivery of CP care
Development of Additively Manufactured Quadrupole Mass Filters for Low-Cost and High-Performance Applications
No full textWith a growing need for more compact and affordable mass spectrometers, many efforts have been made to miniaturize quadrupole mass filters (QMFs). Unfortunately, these efforts have yielded devices with inadequate performance for practical applications in analytical chemistry. This study reports the successful creation of a low-cost, high-performance QMF by means of additive manufacturing. Vat photopolymerization of glass-ceramic feedstock was used to create a novel, monolithic structure, and selective electroless nickel-boron plating metallizes the structure, forming a completed QMF that is lightweight and inexpensive to produce (20 USD per device). Furthermore, additive manufacturing allows QMF dimensions to be rapidly scaled to the optimal sizes for a given application, which is larger than most prior affordable quadrupole designs. Despite the limited precision of additive manufacturing, optimization techniques can be leveraged to produce high-quality devices with smooth surfaces. As a result, our QMFs achieved mass resolutions up to 164 at 69 Da, with abundance sensitivities sufficient to detect carbon-13 isotopes at lower masses—a level of performance comparable to commercial devices. These results indicate that additive manufacturing, properly employed, can significantly advance the state of the art of QMFs and other mass spectrometry technologies.Ph.D
Nanofabrication of silk microneedles for high-throughput micronutrient delivery and continuous sap monitoring in plants
No full textBiomaterials bridging the biotic–abiotic interface in plants offer the opportunity to precisely deliver agrochemicals and continuously monitor plant health, with the goals of increasing resilience to climate change, enhancing crop production and mitigating environmental impact. In this study we report the manipulation of silk fibroin assembly with inorganics nucleation at their phase front to nanomanufacture porous and hollow microneedles that can be interfaced with plants. Plant growth analysis and quantification of wounding gene expression show a non-significant systemic wounding response to the injection of silk microneedles in tomato plants. Microneedles with a hollow structure enable the systemic delivery of plant micronutrients to treat chlorosis in tomato plants and crop biofortification through transport of human micronutrients injected in the petiole and loaded into tomato fruits. Hollow microneedles also provide access to plant vasculature for sap sampling, enabling continuous monitoring and early detection of phytoaccumulation of environmental contaminants such as cadmium
Methods for Enhancing Robustness and Generalization in Machine Learning
No full textWe propose two methods for improving subgroup robustness and out of distribution generalization of machine learning models. First we introduce a formulation of Group DRO with soft group assignment. This formulation can be applied to data with noisy or uncertain group labels, or when only a small subset of the training data has group labels. We propose a modified loss function, explain how to apply it to data with noisy group labels as well as data with missing or few group labels, and perform experiments to demonstrate its effectiveness. In the second part, we propose an invariant decision tree objective that aims to improve the robustness of tree-based models and address a common failure mode of existing methods for out-of-domain generalization. We demonstrate the benefits of this method both theoretically and empirically. Both these approaches are designed to enhance machine learning models’ performance under distribution shift.S.M
Ending Well, Making the Harvest-Paths of Our Values
No full textAny single story shrinks all others. In a place historically cultivated for the cocoa cash crop, this thesis proposes reorienting architectural practice towards a plural valuing of land and its constituent spirits. The journey begins in 2022 with my acquisition of a 99-year lease for a 5-acre land in Ghana. Prior to the conception of an academic proposal, this was to preserve and grow ecological and financial value through time.
Located on a hill-cluster in the Eastern Region, this place is crucial as the birthplace of Ghana’s cocoa industry, which became the world’s largest exporter by 1911. Spurred by economic and colonial incentives, farmer-settlers acquired and cultivated forest land including the one I presently steward. They forged communities that live on despite a subsequent decline of cocoa production in the region. Five centuries of colonial influence in West Africa reduced a plural landscape into singular extractive narratives, creating place-names like the Gold Coast, renamed Ghana after independence. The capitalist framework of monocultural extraction, one reliant on a colonial government and its land survey department, continues under contemporary African states. Architecture and planning—a practice historically tied to power and capital—remains instrumental in this system, often overlooking other ways of valuing land.
This thesis confronts the dispositions of an inherited profession by foregrounding the practices and materials of a socio-cultural paradigm. It is epitomized by the tree called Newbouldia laevis (African boundary tree) and its plural meanings in West Africa. It follows a cocoa harvest-path from a community named after a farmer-settler, Yaa-Aso, and ascends the hills, crossing the land limits of 7 farmers. It ends on the land I hold, with a lease ending in CE 2122.
In July 2024, I led a convocation of the farmers along the path in the defunct cocoa distribution building, toward framing futures based on other values apart from capital. 3 languages were spoken in that gathering - Twi, Anlo-Eʋe and English. It resulted in a 7-foot expansion of the path, and the pacification of a seasonal spirit-stream that crosses it. They set the context for imagining a series of 5 moments, herein recorded, that explore a value system of things spiritual and communal, offered by the transgressions of a widened path and the land I hold at its end.M.Arch
FlexpushdownDB: rethinking computation pushdown for cloud OLAP DBMSs
No full textModern cloud-native OLAP databases adopt a storage-disaggregation architecture that separates the management of computation and storage. A major bottleneck in such an architecture is the network connecting the computation and storage layers. Computation pushdown is a promising solution to tackle this issue, which offloads some computation tasks to the storage layer to reduce network traffic. This paper presents FlexPushdownDB (FPDB), where we revisit the design of computation pushdown in a storage-disaggregation architecture, and then introduce several optimizations to further accelerate query processing. First, FPDB supports hybrid query execution, which combines local computation on cached data and computation pushdown to cloud storage at a fine granularity. Within the cache, FPDB uses a novel Weighted-LFU cache replacement policy that takes into account the cost of pushdown computation. Second, we design adaptive pushdown as a new mechanism to avoid throttling the storage-layer computation during pushdown, which pushes the request back to the computation layer at runtime if the storage-layer computational resource is insufficient. Finally, we derive a general principle to identify pushdown-amenable computational tasks, by summarizing common patterns of pushdown capabilities in existing systems, and further propose two new pushdown operators, namely, selection bitmap and distributed data shuffle. Evaluation on SSB and TPC-H shows each optimization can improve the performance by 2.2 × , 1.9 × , and 3 × respectively
Data-Driven Insights into the Structural Essence of Plasticity in High-Entropy Alloys
No full textThe heterogeneous mechanical response of a crystalline alloy with multiple principal elements was investigated using molecular dynamics simulations. The local configuration of the alloy in its quiescent state was characterized by the variables derived from the gyration tensor and the atomic electronegativity. A multivariate analysis identified the geometric and chemical factors that influenced the atomic packing variations. Upon straining, the non-affine displacement exhibited spatial heterogeneity. A statistical correlation was established between the local yield events and the specific features of the local configuration. Our findings, validated by the performance metrics analysis, provided a structural criterion for the instability mechanisms in high-entropy alloys (HEAs) and enhanced the understanding of their plasticity