150813 research outputs found
Sort by
History of chemically and radiatively important atmospheric gases from the Advanced Global Atmospheric Gases Experiment (AGAGE)
No full textWe present the organization, instrumentation, datasets, data interpretation, modeling, and accomplishments of the multinational global atmospheric measurement program AGAGE (Advanced Global Atmospheric Gases Experiment). AGAGE is distinguished by its capability to measure globally, at high frequency, and at multiple sites all the important species in the Montreal Protocol and all the important non-carbon-dioxide (non-CO2) gases assessed by the Intergovernmental Panel on Climate Change (CO2 is also measured at several sites). The scientific objectives of AGAGE are important in furthering our understanding of global chemical and climatic phenomena. They are the following: (1) to accurately measure the temporal and spatial distributions of anthropogenic gases that contribute the majority of reactive halogen to the stratosphere and/or are strong infrared absorbers (chlorocarbons, chlorofluorocarbons – CFCs, bromocarbons, hydrochlorofluorocarbons – HCFCs, hydrofluorocarbons – HFCs and polyfluorinated compounds (perfluorocarbons – PFCs), nitrogen trifluoride – NF3, sulfuryl fluoride – SO2F2, and sulfur hexafluoride – SF6) and use these measurements to determine the global rates of their emission and/or destruction (i.e., lifetimes); (2) to accurately measure the global distributions and temporal behaviors and determine the sources and sinks of non-CO2 biogenic–anthropogenic gases important to climate change and/or ozone depletion (methane – CH4, nitrous oxide – N2O, carbon monoxide – CO, molecular hydrogen – H2, methyl chloride – CH3Cl, and methyl bromide – CH3Br); (3) to identify new long-lived greenhouse and ozone-depleting gases (e.g., SO2F2, NF3, heavy PFCs (C4F10, C5F12, C6F14, C7F16, and C8F18) and hydrofluoroolefins (HFOs; e.g., CH2 = CFCF3) have been identified in AGAGE), initiate the real-time monitoring of these new gases, and reconstruct their past histories from AGAGE, air archive, and firn air measurements; (4) to determine the average concentrations and trends of tropospheric hydroxyl radicals (OH) from the rates of destruction of atmospheric trichloroethane (CH3CCl3), HFCs, and HCFCs and estimates of their emissions; (5) to determine from atmospheric observations and estimates of their destruction rates the magnitudes and distributions by region of surface sources and sinks of all measured gases; (6) to provide accurate data on the global accumulation of many of these trace gases that are used to test the synoptic-, regional-, and global-scale circulations predicted by three-dimensional models; and (7) to provide global and regional measurements of methane, carbon monoxide, and molecular hydrogen and estimates of hydroxyl levels to test primary atmospheric oxidation pathways at midlatitudes and the tropics
Computational Modeling of Biological Function
No full textHow biological function emerges from complex molecular patterns is a fundamental question in biology. Addressing this question requires a deep exploration of the concepts of genotype and phenotype, which serve as the foundation of this inquiry. This dissertation focuses on providing a quantitative approach through the lens of computation to dissect the dynamic relationship between genotype and phenotype. In particular, recent advancements in high-content genotyping methods, such as genome-wide association studies (GWAS) and single-cell RNA sequencing, have provided powerful tools for mapping the molecular basis of biological function, but also have introduced challenges due to the high dimensionality, vast combinatorial possibilities, and multimodal characteristics of the data. The overarching goal of this dissertation is first to provide a critical discussion on the theories of genotype and phenotype as they relate to biological function and propose new methods to map their relationship. Specifically, we present the integrated genetics framework designed to analyze and interpret the manifold of genotypes and their associated phenotypes simultaneously. We applied this approach to develop a multimodal foundation model for human transcriptomics at the cellular level. To further test the capabilities of this method, we apply it to dissect the aging process. The results of this study provide novel concepts and methods for analyzing the genetic data along with phenotypic information with higher resolution. Moreover, the results shed light on uncovered potential cross-tissue biomarkers that are undetectable through conventional gene expression analysis alone. Overall, this study aims to advance our understanding of the dynamic interplay between gene patterns and phenotypic manifestation and demonstrates the potential of computational modeling in uncovering new dimensions of cellular function and complexity.Ph.D
A Sublinear Algorithm for Approximate Shortest Paths in Large Networks
No full textWSDM ’25, March 10–14, 2025, Hannover, GermanyComputing distances and finding shortest paths in massive real-world networks is a fundamental algorithmic task in network analysis. There are two main approaches to solving this task. On one end are traversal-based algorithms like bidirectional breadth-first search (BiBFS), which have no preprocessing step but are slow on individual distance inquiries. On the other end are indexing-based approaches, which create and maintain a large index. This allows for answering individual inquiries very fast; however, index creation is prohibitively expensive. We seek to bridge these two extremes: quickly answer distance inquiries without the need for costly preprocessing.
We propose a new algorithm and data structure, WormHole, for approximate shortest path computations. WormHole leverages structural properties of social networks to build a sublinearly sized index, drawing upon the core-periphery decomposition of Ben-Eliezer et al. [10]. Empirically, WormHole's preprocessing time improves upon index-based solutions by orders of magnitude: indexing billion edges graphs takes only a few minutes. Real time performance is consistently much faster than in BiBFS. The acceleration comes at the cost of a minor accuracy trade-off. We complement these empirical results with provable theoretical guarantees
Transformers as Empirical Bayes Estimators The Poisson Model
No full textWe study the ability of transformers to perform In Context Learning (ICL) in the setting of Empirical Bayes for the Poison Model. On the theoretical side, we demonstrate the expressibility of transformers by formulating a way to approximate the Robbins estimator, the first empirical Bayes estimator for the Poisson model. On the empirical side, we show that transformers pre-trained on synthetic data can generalize to unseen prior and sequence lengths, outperforming existing methods like Robbins, NPMLE, and ERM monotone in efficiency and accuracy. By studying the internal behavior of the representations of the intermediate layers of these transformers, we found that the representation converges quickly and smoothly over the layers. We also demonstrate that it’s unlikely transformers are implementing Robbin’s or NPMLE estimators in context.M.Eng
Strange Attitudes on Top
No full textThis dissertation investigates how attitude verbs of belief and desire engage with embedded material of a similar nature. Chapter 1 looks at the (cross-linguistically unusual) Slovenian existential doxastic attitude verb dopuščati (‘allow for the possibility’) and the embedding of epistemic modal verbs under it. Chapter 2 looks at the (overall puzzling) want and its Slovenian counterpart hoteti, and at their behaviour with respect to embedded doxastic attitudes, epistemic adverbs, and epistemic adjectives. Chapter 3 looks at the (cross-linguistically unusual) Koryak variable-force variable-flavour attitude verb ivək (‘think’, ‘allow for the possibility’, ‘say’, ‘suggest’) and at how its apparent bouletic flavour (‘wish’, ‘hope’, ‘fear’) is derived with the help of covert desiderative components inside the embedded clause. Attitude verbs have the standard role as quantifiers over possible worlds (Hintikka 1962), parameters of evaluation are assumed to contain a set of worlds called the information state (Yalcin 2007; a.o.), which the attitude verb modifies and passes to the embedded clause, while the epistemic modal base is taken to be ‘local’, forming a subset of the information state (Mandelkern 2017, 2019a). Some of the overarching theoretical contributions are the introduction of a new parameter of evaluation (‘selected state’), which is crucial in modelling embedding under non-universal attitude verbs, and a refined view of epistemic modality. Subjective epistemic modality is proposed to involve a second constraint on the shape of the modal base, whose effect is to strengthen embedded necessity claims and help derive the infelicities observed in chapters 1 and 2. We also address the connection between beliefs and desires in the context of various desire interpretations (wants in chapter 2, hopes and wishes in chapter 3).Ph.D
Interactive Spin Dynamics in Magnon and Quantum Spin Systems
No full textSpintronics utilizes the intrinsic spin of electrons to design next-generation electronic devices, reducing power consumption and enabling innovative computing functions. Over the past decades, significant research interest has been directed toward two types of spin-based systems: collective excitations of spins, known as spin waves or magnons, in magnetic materials, and optically active spin defects as represented by nitrogen-vacancy (NV) centers in diamond, leading to the prosperity of magnonics, quantum sensing, and quantum information processing. As the understanding of dynamics in individual spin systems has deepened, recently there has been an increasing interest in the interactive dynamics within hybrid spin systems. This shift in focus reflects an increasing curiosity about how these complex interactions can be harnessed to further advance their microwave and quantum applications. However, several challenges persist, including the limited coherence length of magnons and the restricted frequency range of NV-based magnetometers, which will be tackled in this thesis. We first leverage the chirality of interlayer magnetic dipolar interactions to introduce an easily implementable system—antiparallel aligned magnetic multilayers—for realizing topological magnonic surface states and low-dissipation spin current transport in a tunable manner. We then expand the frequency window of NV-based magnetometers using nonlinear microwave-spin interactions, offering novel functionalities in quantum state control and sensing. We further exploit nonlinear spin dynamics by hybridizing NV centers with magnonic thin films, which not only amplifies the intensity of nonlinear resonance signals that are intrinsic to NV spins, but also enables novel frequency mixings through parametric pumping and nonlinear magnon scattering effects. We believe our study of interactive spin dynamics in hybrid systems involving magnons, quantum spin defects, and microwave photons help optimize these systems for a wide range of applications in both classical and quantum domains.Ph.D
Demand Forecasting Analysis for Pharma Retail
No full textDemand planning is the connection between marketing, finance, and operations. In an industry like pharma retail, products do not always behave according to a regular stable baseline. In addition, marketing enrichment like promotions or price fluctuations and the impactof government regulations and patient base characteristics increase operational complexity. Moreover, more than thirty percent of changes in the forecast from one cycle to another can lead to overstock or out-of-stock due to the high production and delivery lead times.
The purpose of this project is to find a proper demand forecasting model for a selected group of stock-keeping units to improve supply processes of the most important stores of the sponsoring company, leading to further benefits such as budget purposes as a top-down analysis. Data analysis is needed for trends, seasonality, stockouts, and demand stability. Followed by the application of various forecasting models, including Machine Learning algorithms, this project provides a comparison of models to define the best baseline as a tool for the planning area to enrich to improve operational KPIs
GraphPipe: Improving Performance and Scalability of DNN Training with Graph Pipeline Parallelism
No full textASPLOS ’25, March 30–April 3, 2025, Rotterdam, NetherlandsDeep neural networks (DNNs) continue to grow rapidly in size, making them infeasible to train on a single device (e.g. GPU). Pipeline parallelism is commonly used in existing DNN systems to support large-scale DNN training by partitioning a DNN into multiple stages, which concurrently perform DNN computation for different micro-batches of training samples in a pipeline fashion. However, existing pipeline-parallel approaches only consider sequential pipeline stages and thus ignore the topology of a DNN, resulting in missed model-parallel opportunities.
This paper presents graph pipeline parallelism (GPP), a new pipeline-parallel scheme that partitions a DNN into pipeline stages whose dependencies are identified by a directed acyclic graph. GPP generalizes existing sequential pipeline parallelism and preserves the inherent topology of a DNN to enable concurrent execution of computationally-independent operators, resulting in reduced memory requirement and improved GPU performance. In addition, we develop GraphPipe, a distributed system that exploits GPP strategies to enable performant and scalable DNN training. GraphPipe partitions a DNN into a graph of stages, optimizes micro-batch schedules for these stages, and parallelizes DNN training using the discovered GPP strategies. Evaluation on a variety of DNNs shows that GraphPipe outperforms existing pipeline-parallel systems such as PipeDream and Piper by up to 1.6×. GraphPipe also reduces the search time by 9-21× compared to PipeDream and Piper
Revisiting Wireless Cyberattacks on Vehicles
No full textThe automotive industry has been a prime target for cybercriminals for decades, with attacks becoming more sophisticated as vehicles integrate advanced digital technologies. In response, new standards and regulations have been introduced, requiring manufacturers to implement robust cybersecurity measures to obtain necessary certifications. Modern vehicles have an extensive attack surface due to the increasing number of interconnected electronic components and wireless communication features. While new technologies improve connectivity, automation, and comfort, they also introduce new vulnerabilities that can be exploited by attackers. This paper presents a comprehensive analysis of the attack surface of modern vehicles, focusing on the security risks associated with wireless communication technologies. Each technology is examined in detail, highlighting existing research, known vulnerabilities, and potential countermeasures. Furthermore, this study identifies key research gaps in the field, providing insights into critical areas that require further investigation. This work aims to guide future research efforts in order to enhance vehicle cybersecurity in the evolving landscape of smart, autonomous, and connected vehicles
Superparamagnetic Tunnel Junctions for Reliable True Randomness and Efficient Probabilistic Machine Learning
No full textPhysical devices exhibiting stochastic functions with low energy consumption and high device density have the potential to enable complex probability-based computing algorithms, accelerate machine learning tasks, and enhance hardware security. Recently, superparamagnetic tunnel junctions (sMTJs) have been widely explored for such purposes, leading to the development of limited-scale sMTJ-based systems. Existing sMTJs face significant scalability and reliability issues, however, because their intrinsically low energy barrier and correspondingly small device area result in high sensitivity to external perturbations, as well as large variations from device to device. Here, we present an experimental demonstration of three-terminal sMTJs as reliable and potentially scalable sources of true randomness in the field-free regime. By leveraging dual-current controllability and incorporating feedback, we stabilize the switching operation of superparamagnets and reach cryptographic-quality random bitstreams. The realization of controllable and robust true random sMTJs underpin a general hardware platform for computing schemes exploiting the stochasticity in the physical world, as demonstrated by the generative artificial intelligence example in our experiment. Furthermore, we experimentally demonstrate a novel method of utilizing sMTJs as stochastic analog-to-digital converters (sADCs) in a crossbar array architecture for neural network acceleration, showing performance comparable to software implementations. This work highlights the potential of sMTJs to revolutionize energy-efficient computing and provides a foundation for future advancements in probabilistic computing and hardware security.S.M