Showing papers by "Lehigh University published in 2018"
••
TL;DR: The authors provides a review and commentary on the past, present, and future of numerical optimization algorithms in the context of machine learning applications and discusses how optimization problems arise in machine learning and what makes them challenging.
Abstract: This paper provides a review and commentary on the past, present, and future of numerical optimization algorithms in the context of machine learning applications. Through case studies on text classification and the training of deep neural networks, we discuss how optimization problems arise in machine learning and what makes them challenging. A major theme of our study is that large-scale machine learning represents a distinctive setting in which the stochastic gradient (SG) method has traditionally played a central role while conventional gradient-based nonlinear optimization techniques typically falter. Based on this viewpoint, we present a comprehensive theory of a straightforward, yet versatile SG algorithm, discuss its practical behavior, and highlight opportunities for designing algorithms with improved performance. This leads to a discussion about the next generation of optimization methods for large-scale machine learning, including an investigation of two main streams of research on techniques th...
2,238 citations
••
18 Jun 2018
TL;DR: AttnGAN as mentioned in this paper proposes an attentional generative network to synthesize fine-grained details at different sub-regions of the image by paying attentions to the relevant words in the natural language description.
Abstract: In this paper, we propose an Attentional Generative Adversarial Network (AttnGAN) that allows attention-driven, multi-stage refinement for fine-grained text-to-image generation. With a novel attentional generative network, the AttnGAN can synthesize fine-grained details at different sub-regions of the image by paying attentions to the relevant words in the natural language description. In addition, a deep attentional multimodal similarity model is proposed to compute a fine-grained image-text matching loss for training the generator. The proposed AttnGAN significantly outperforms the previous state of the art, boosting the best reported inception score by 14.14% on the CUB dataset and 170.25% on the more challenging COCO dataset. A detailed analysis is also performed by visualizing the attention layers of the AttnGAN. It for the first time shows that the layered attentional GAN is able to automatically select the condition at the word level for generating different parts of the image.
1,217 citations
••
TL;DR: This paper provides a latest survey of the physical layer security research on various promising 5G technologies, includingPhysical layer security coding, massive multiple-input multiple-output, millimeter wave communications, heterogeneous networks, non-orthogonal multiple access, full duplex technology, and so on.
Abstract: Physical layer security which safeguards data confidentiality based on the information-theoretic approaches has received significant research interest recently. The key idea behind physical layer security is to utilize the intrinsic randomness of the transmission channel to guarantee the security in physical layer. The evolution toward 5G wireless communications poses new challenges for physical layer security research. This paper provides a latest survey of the physical layer security research on various promising 5G technologies, including physical layer security coding, massive multiple-input multiple-output, millimeter wave communications, heterogeneous networks, non-orthogonal multiple access, full duplex technology, and so on. Technical challenges which remain unresolved at the time of writing are summarized and the future trends of physical layer security in 5G and beyond are discussed.
580 citations
••
Fisk University1, Vanderbilt University2, Lehigh University3, Harvard University4, Northern Kentucky University5, University of Toulouse6, University of California, Berkeley7, University of California, Riverside8, Georgia State University9, University of North Carolina at Chapel Hill10, Boston University11, Andrés Bello National University12, INAF13, Space Telescope Science Institute14, George Mason University15
TL;DR: In this paper, an erratum for this article has been published in 2018 AJ 156 183, https://doi.org/10.3847/1538-3881/aade86
Abstract: An erratum for this article has been published in 2018 AJ 156 183, https://doi.org/10.3847/1538-3881/aade86
487 citations
••
TL;DR: A novel simulation method to determine thermodynamic phase diagrams as a function of the total protein concentration and temperature is developed and is capable of capturing qualitative changes in the phase diagram due to phosphomimetic mutations of FUS and to the presence or absence of the large folded domain in LAF-1.
Abstract: Membraneless organelles important to intracellular compartmentalization have recently been shown to comprise assemblies of proteins which undergo liquid-liquid phase separation (LLPS). However, many proteins involved in this phase separation are at least partially disordered. The molecular mechanism and the sequence determinants of this process are challenging to determine experimentally owing to the disordered nature of the assemblies, motivating the use of theoretical and simulation methods. This work advances a computational framework for conducting simulations of LLPS with residue-level detail, and allows for the determination of phase diagrams and coexistence densities of proteins in the two phases. The model includes a short-range contact potential as well as a simplified treatment of electrostatic energy. Interaction parameters are optimized against experimentally determined radius of gyration data for multiple unfolded or intrinsically disordered proteins (IDPs). These models are applied to two systems which undergo LLPS: the low complexity domain of the RNA-binding protein FUS and the DEAD-box helicase protein LAF-1. We develop a novel simulation method to determine thermodynamic phase diagrams as a function of the total protein concentration and temperature. We show that the model is capable of capturing qualitative changes in the phase diagram due to phosphomimetic mutations of FUS and to the presence or absence of the large folded domain in LAF-1. We also explore the effects of chain-length, or multivalency, on the phase diagram, and obtain results consistent with Flory-Huggins theory for polymers. Most importantly, the methodology presented here is flexible so that it can be easily extended to other pair potentials, be used with other enhanced sampling methods, and may incorporate additional features for biological systems of interest.
381 citations
••
TL;DR: A novel end-to-end adversarial critic network with a multi-scale L1 loss function to force the critic and segmentor to learn both global and local features that capture long- and short-range spatial relationships between pixels is proposed.
Abstract: Inspired by classic Generative Adversarial Networks (GANs), we propose a novel end-to-end adversarial neural network, called SegAN, for the task of medical image segmentation. Since image segmentation requires dense, pixel-level labeling, the single scalar real/fake output of a classic GAN's discriminator may be ineffective in producing stable and sufficient gradient feedback to the networks. Instead, we use a fully convolutional neural network as the segmentor to generate segmentation label maps, and propose a novel adversarial critic network with a multi-scale L1 loss function to force the critic and segmentor to learn both global and local features that capture long- and short-range spatial relationships between pixels. In our SegAN framework, the segmentor and critic networks are trained in an alternating fashion in a min-max game: The critic is trained by maximizing a multi-scale loss function, while the segmentor is trained with only gradients passed along by the critic, with the aim to minimize the multi-scale loss function. We show that such a SegAN framework is more effective and stable for the segmentation task, and it leads to better performance than the state-of-the-art U-net segmentation method. We tested our SegAN method using datasets from the MICCAI BRATS brain tumor segmentation challenge. Extensive experimental results demonstrate the effectiveness of the proposed SegAN with multi-scale loss: on BRATS 2013 SegAN gives performance comparable to the state-of-the-art for whole tumor and tumor core segmentation while achieves better precision and sensitivity for Gd-enhance tumor core segmentation; on BRATS 2015 SegAN achieves better performance than the state-of-the-art in both dice score and precision.
376 citations
••
TL;DR: It is argued that although automated text analysis cannot be used to study all phenomena, it is a useful tool for examining patterns in text that neither researchers nor consumers can detect unaided.
Abstract: The amount of digital text available for analysis by consumer researchers has risen dramatically. Consumer discussions on the internet, product reviews, and digital archives of news articles and press releases are just a few potential sources for insights about consumer attitudes, interaction, and culture. Drawing from linguistic theory and methods, this article presents an overview of automated text analysis, providing integration of linguistic theory with constructs commonly used in consumer research, guidance for choosing amongst methods, and advice for resolving sampling and statistical issues unique to text analysis. We argue that although automated text analysis cannot be used to study all phenomena, it is a useful tool for examining patterns in text that neither researchers nor consumers can detect unaided. Text analysis can be used to examine psychological and sociological constructs in consumer-produced digital text by enabling discovery or by providing ecological validity.
359 citations
••
28 Feb 2018TL;DR: In this paper, a unique set of well-defined silica-supported Ni nanoclusters (1-7 nm) and advanced characterization methods were used to prove how structure sensitivity influences the mechanism of catalytic CO2 reduction.
Abstract: Continuous efforts in the field of materials science have allowed us to generate smaller and smaller metal nanoparticles, creating new opportunities to understand catalytic properties that depend on the metal particle size. Structure sensitivity is the phenomenon where not all surface atoms in a supported metal catalyst have the same activity. Understanding structure sensitivity can assist in the rational design of catalysts, allowing control over mechanisms, activity and selectivity, and thus even the viability of a catalytic reaction. Here, using a unique set of well-defined silica-supported Ni nanoclusters (1–7 nm) and advanced characterization methods, we prove how structure sensitivity influences the mechanism of catalytic CO2 reduction, the nature of which has been long debated. These findings bring fundamental new understanding of CO2 hydrogenation over Ni and allow us to control both activity and selectivity, which can be a means for CO2 emission abatement through its valorization as a low- or even negative-cost feedstock on a low-cost transition-metal catalyst.
337 citations
••
TL;DR: The supported V2O5-WO3/TiO2 catalysts have become the most widely used industrial catalysts for selective catalytic reduction (SCR) applications since introduction of this technology in the early 1970s as mentioned in this paper.
Abstract: The selective catalytic reduction (SCR) of NOx with NH3 to harmless N2 and H2O plays a crucial role in reducing highly undesirable NOx acid gas emissions from large utility boilers, industrial boilers, municipal waste plants, and incinerators. The supported V2O5–WO3/TiO2 catalysts have become the most widely used industrial catalysts for these SCR applications since introduction of this technology in the early 1970s. This Perspective examines the current fundamental understanding and recent advances of the supported V2O5–WO3/TiO2 catalyst system: (i) catalyst synthesis, (ii) molecular structures of titania-supported vanadium and tungsten oxide species, (iii) surface acidity, (iv) catalytic active sites, (v) surface reaction intermediates, (vi) reaction mechanism, (vii) rate-determining-step, and (viii) reaction kinetics.
315 citations
•
03 Dec 2018TL;DR: This work presents an end-to-end framework for solving the Vehicle Routing Problem (VRP) using reinforcement learning, and demonstrates how this approach can handle problems with split delivery and explore the effect of such deliveries on the solution quality.
Abstract: We present an end-to-end framework for solving the Vehicle Routing Problem (VRP) using reinforcement learning. In this approach, we train a single policy model that finds near-optimal solutions for a broad range of problem instances of similar size, only by observing the reward signals and following feasibility rules. We consider a parameterized stochastic policy, and by applying a policy gradient algorithm to optimize its parameters, the trained model produces the solution as a sequence of consecutive actions in real time, without the need to re-train for every new problem instance. On capacitated VRP, our approach outperforms classical heuristics and Google's OR-Tools on medium-sized instances in solution quality with comparable computation time (after training). We demonstrate how our approach can handle problems with split delivery and explore the effect of such deliveries on the solution quality. Our proposed framework can be applied to other variants of the VRP such as the stochastic VRP, and has the potential to be applied more generally to combinatorial optimization problems.
301 citations
••
TL;DR: A systematic review of the SR-based multi-sensor image fusion literature, highlighting the pros and cons of each category of approaches and evaluating the impact of these three algorithmic components on the fusion performance when dealing with different applications.
••
TL;DR: In this article, the authors present estimates of how many exoplanets the Transiting Exoplanet Survey Satellite (TESS) will detect, the physical properties of the detected planets, and the properties of those planets that those planets orbit.
Abstract: The Transiting Exoplanet Survey Satellite (TESS) has a goal of detecting small planets orbiting stars bright enough for mass determination via ground-based radial velocity observations. Here, we present estimates of how many exoplanets the TESS mission will detect, the physical properties of the detected planets, and the properties of the stars that those planets orbit. This work uses stars drawn from the TESS Input Catalog Candidate Target List and revises yields from prior studies that were based on Galactic models. We modeled the TESS observing strategy to select approximately 200,000 stars at 2-minute cadence, while the remaining stars are observed at 30-minute cadence in full-frame image data. We placed zero or more planets in orbit around each star, with physical properties following measured exoplanet occurrence rates, and used the TESS noise model to predict the derived properties of the detected exoplanets. In the TESS 2-minute cadence mode we estimate that TESS will find 1250 ± 70 exoplanets (90% confidence), including 250 smaller than 2 R(sub ⊕). Furthermore, we predict that an additional 3100 planets will be found in full-frame image data orbiting bright dwarf stars and more than 10,000 around fainter stars. We predict that TESS will find 500 planets orbiting M dwarfs, but the majority of planets will orbit stars larger than the Sun. Our simulated sample of planets contains hundreds of small planets amenable to radial velocity follow-up, potentially more than tripling the number of planets smaller than 4 R(sub ⊕) with mass measurements. This sample of simulated planets is available for use in planning follow-up observations and analyses.
••
TL;DR: A unified structural view of hnRNPA2 self-assembly, aggregation, and interaction and the distinct effects of small chemical changes-disease mutations and arginine methylation-on these assemblies are provided.
••
TL;DR: The results show that >50% of metro lines are highly exposed to flood risk, indicating that the Guangzhou metro system is vulnerable to flood events.
••
TL;DR: The results show that these characteristic temperatures are highly correlated, suggesting that experiments performed in dilute conditions may be used to predict phase separation, and suggest that smaller simulations or experiments to determine Tθ or TB can provide useful insights into the corresponding phase behavior.
Abstract: Proteins that undergo liquid-liquid phase separation (LLPS) have been shown to play a critical role in many physiological functions through formation of condensed liquid-like assemblies that function as membraneless organelles within biological systems. To understand how different proteins may contribute differently to these assemblies and their functions, it is important to understand the molecular driving forces of phase separation and characterize their phase boundaries and material properties. Experimental studies have shown that intrinsically disordered regions of these proteins are a major driving force, as many of them undergo LLPS in isolation. Previous work on polymer solution phase behavior suggests a potential correspondence between intramolecular and intermolecular interactions that can be leveraged to discover relationships between single-molecule properties and phase boundaries. Here, we take advantage of a recently developed coarse-grained framework to calculate the θ temperature [Formula: see text], the Boyle temperature [Formula: see text], and the critical temperature [Formula: see text] for 20 diverse protein sequences, and we show that these three properties are highly correlated. We also highlight that these correlations are not specific to our model or simulation methodology by comparing between different pairwise potentials and with data from other work. We, therefore, suggest that smaller simulations or experiments to determine [Formula: see text] or [Formula: see text] can provide useful insights into the corresponding phase behavior.
••
TL;DR: A systematical survey of the state-of-the-art caching techniques that were recently developed in cellular networks, including macro-cellular networks, heterogeneous networks, device-to-device networks, cloud-radio access networks, and fog-radioaccess networks.
Abstract: Mobile data traffic is currently growing exponentially and these rapid increases have caused the backhaul data rate requirements to become the major bottleneck to reducing costs and raising revenue for operators. To address this problem, caching techniques have attracted significant attention since they can effectively reduce the backhaul traffic by eliminating duplicate data transmission that carries popular content. In addition, other system performance metrics can also be improved through caching techniques, e.g., spectrum efficiency, energy efficiency, and transmission delay. In this paper, we provide a systematical survey of the state-of-the-art caching techniques that were recently developed in cellular networks, including macro-cellular networks, heterogeneous networks, device-to-device networks, cloud-radio access networks, and fog-radio access networks. In particular, we give a tutorial on the fundamental caching techniques and introduce caching algorithms from three aspects, i.e., content placement, content delivery, and joint placement and delivery. We provide comprehensive comparisons among different algorithms in terms of different performance metrics, including throughput, backhaul cost, power consumption, and network delay. Finally, we summarize the main research achievements in different networks, and highlight main challenges and potential research directions.
••
Massachusetts Institute of Technology1, University of Texas at Austin2, Princeton University3, University of Southern Queensland4, University of California, Riverside5, Harvard University6, Ames Research Center7, Search for extraterrestrial intelligence8, University of Geneva9, NASA Exoplanet Science Institute10, University of Maryland, College Park11, Johns Hopkins University12, Lehigh University13, Eötvös Loránd University14, Hungarian Academy of Sciences15, Vanderbilt University16, INAF17, Aarhus University18, Yale University19, Goddard Space Flight Center20, University of Chicago21, Technical University of Denmark22, Konkoly Thege Miklós Astronomical Institute23, Tokyo Institute of Technology24, Cornell University25, Spanish National Research Council26, Carnegie Institution for Science27, University of Florida28
TL;DR: The detection of a transiting planet around π Men (HD 39091), using data from the Transiting Exoplanet Survey Satellite (TESS), is reported, confirming the existence of the planet and leading to a mass determination of 4.82±0.85 M ⊕.
Abstract: We report the detection of a transiting planet around π Men (HD 39091), using data from the Transiting Exoplanet Survey Satellite (TESS). The solar-type host star is unusually bright (V = 5.7) and was already known to host a Jovian planet on a highly eccentric, 5.7-year orbit. The newly discovered planet has a size of 2.04 ± 0.05 R⊕ and an orbital period of 6.27 days. Radial-velocity data from the HARPS and AAT/UCLES archives also displays a 6.27-day periodicity, confirming the existence of the planet and leading to a mass determination of 4.82±0.85 M⊕. The star's proximity and brightness will facilitate further investigations, such as atmospheric spectroscopy, asteroseismology, the Rossiter-McLaughlin effect, astrometry, and direct imaging.
••
University of Wisconsin-Madison1, University of Minnesota2, University of California, Merced3, Drexel University4, University of Oregon5, Pennsylvania State University6, Kent State University7, North Dakota State University8, United States Geological Survey9, Lehigh University10, Umeå University11, University of Illinois at Urbana–Champaign12, University of Göttingen13, University of Arizona14, University of Chile15, Columbia University16, Academy of Natural Sciences of Drexel University17, Kyoto Prefectural University18
TL;DR: The Neotoma Paleoecology Database as mentioned in this paper is a community-curated data resource that supports interdisciplinary global change research by enabling broad-scale studies of taxon and community diversity, distr...
••
University of Exeter1, University of Utah2, University of Leicester3, Imperial College London4, University of Hawaii at Manoa5, Lund University6, Russian Academy of Sciences7, Lehigh University8, University of Wisconsin–La Crosse9, Bowdoin College10, Uppsala University11, University of York12, University of Toulouse13, Adam Mickiewicz University in Poznań14, University of Toronto15, Université du Québec à Montréal16, University of California, Los Angeles17, University of Southampton18, Met Office19, United States Geological Survey20, University of Tartu21, University of Alaska Anchorage22, University of Helsinki23, University of Victoria24, University of Nottingham25, Laval University26, Texas A&M University27, Newcastle University28, Geological Survey of Finland29, Oeschger Centre for Climate Change Research30, University of Santiago de Compostela31, University of Aberdeen32, Trinity College, Dublin33, University of Queensland34, Lamont–Doherty Earth Observatory35, University of Lapland36, Norwegian Polar Institute37, Ontario Ministry of Natural Resources38, Champlain College39, Stockholm University40, University of Leeds41, Forestry Commission42, University of Amsterdam43, Chinese Academy of Sciences44, Northeast Normal University45
TL;DR: This article examined the global relationship between peatland carbon accumulation rates during the last millennium and planetary-scale climate space and found a positive relationship between carbon accumulation and cumulative photosynthetically active radiation during the growing season for mid-to high-latitude peatlands in both hemispheres.
Abstract: The carbon sink potential of peatlands depends on the balance of carbon uptake by plants and microbial decomposition The rates of both these processes will increase with warming but it remains unclear which will dominate the global peatland response Here we examine the global relationship between peatland carbon accumulation rates during the last millennium and planetary-scale climate space A positive relationship is found between carbon accumulation and cumulative photosynthetically active radiation during the growing season for mid- to high-latitude peatlands in both hemispheres However, this relationship reverses at lower latitudes, suggesting that carbon accumulation is lower under the warmest climate regimes Projections under Representative Concentration Pathway (RCP)26 and RCP85 scenarios indicate that the present-day global sink will increase slightly until around ad 2100 but decline thereafter Peatlands will remain a carbon sink in the future, but their response to warming switches from a negative to a positive climate feedback (decreased carbon sink with warming) at the end of the twenty-first century
••
University of Bern1, Oeschger Centre for Climate Change Research2, University of New South Wales3, Oregon State University4, Australian National University5, University of Cambridge6, Max Planck Society7, British Antarctic Survey8, University of Copenhagen9, University of Bordeaux10, Lamont–Doherty Earth Observatory11, University of Bremen12, University of Toronto13, Durham University14, Leibniz Center for Tropical Marine Ecology15, Alfred Wegener Institute for Polar and Marine Research16, University of Geneva17, Ca' Foscari University of Venice18, Siberian Federal University19, GNS Science20, Queen's University21, University of Lausanne22, Université du Québec à Montréal23, University of Southern California24, University of Nebraska–Lincoln25, University of Bristol26, Adam Mickiewicz University in Poznań27, Yale University28, Université Paris-Saclay29, National Center for Atmospheric Research30, Bjerknes Centre for Climate Research31, École pratique des hautes études32, United States Global Change Research Program33, University of Kiel34, Lund University35, Chinese Academy of Sciences36, Northeast Normal University37, Lehigh University38, Utrecht University39, ETH Zurich40, Peking University41
TL;DR: In this article, an observation-based synthesis of the understanding of past intervals with temperatures within the range of projected future warming suggests that there is a low risk of runaway greenhouse gas feedbacks for global warming of no more than 2 °C.
Abstract: Over the past 3.5 million years, there have been several intervals when climate conditions were warmer than during the pre-industrial Holocene. Although past intervals of warming were forced differently than future anthropogenic change, such periods can provide insights into potential future climate impacts and ecosystem feedbacks, especially over centennial-to-millennial timescales that are often not covered by climate model simulations. Our observation-based synthesis of the understanding of past intervals with temperatures within the range of projected future warming suggests that there is a low risk of runaway greenhouse gas feedbacks for global warming of no more than 2 °C. However, substantial regional environmental impacts can occur. A global average warming of 1–2 °C with strong polar amplification has, in the past, been accompanied by significant shifts in climate zones and the spatial distribution of land and ocean ecosystems. Sustained warming at this level has also led to substantial reductions of the Greenland and Antarctic ice sheets, with sea-level increases of at least several metres on millennial timescales. Comparison of palaeo observations with climate model results suggests that, due to the lack of certain feedback processes, model-based climate projections may underestimate long-term warming in response to future radiative forcing by as much as a factor of two, and thus may also underestimate centennial-to-millennial-scale sea-level rise.
••
TL;DR: A label-free mid-infrared biosensor capable of distinguishing multiple analytes in heterogeneous biological samples with high sensitivity that opens up exciting possibilities for gaining new insights into biological processes such as signaling or transport in basic research as well as provides a valuable toolkit for bioanalytical and pharmaceutical applications.
Abstract: A multitude of biological processes are enabled by complex interactions between lipid membranes and proteins. To understand such dynamic processes, it is crucial to differentiate the constituent biomolecular species and track their individual time evolution without invasive labels. Here, we present a label-free mid-infrared biosensor capable of distinguishing multiple analytes in heterogeneous biological samples with high sensitivity. Our technology leverages a multi-resonant metasurface to simultaneously enhance the different vibrational fingerprints of multiple biomolecules. By providing up to 1000-fold near-field intensity enhancement over both amide and methylene bands, our sensor resolves the interactions of lipid membranes with different polypeptides in real time. Significantly, we demonstrate that our label-free chemically specific sensor can analyze peptide-induced neurotransmitter cargo release from synaptic vesicle mimics. Our sensor opens up exciting possibilities for gaining new insights into biological processes such as signaling or transport in basic research as well as provides a valuable toolkit for bioanalytical and pharmaceutical applications.
••
TL;DR: In this article, the authors reflect on what the experimental measurements have taught us so far, the limitations of the techniques used for studying jets, how the techniques can be improved, and how to move forward with the wealth of experimental data such that a complete description of energy loss in the QGP can be achieved.
Abstract: A hot, dense medium called a quark gluon plasma (QGP) is created in ultrarelativistic heavy ion collisions. Early in the collision, hard parton scatterings generate high momentum partons that traverse the medium, which then fragment into sprays of particles called jets. Understanding how these partons interact with the QGP and fragment into final state particles provides critical insight into quantum chromodynamics. Experimental measurements from high momentum hadrons, two particle correlations, and full jet reconstruction at the Relativistic Heavy Ion Collider (RHIC) and the Large Hadron Collider (LHC) continue to improve our understanding of energy loss in the QGP. Run 2 at the LHC recently began and there is a jet detector at RHIC under development. Now is the perfect time to reflect on what the experimental measurements have taught us so far, the limitations of the techniques used for studying jets, how the techniques can be improved, and how to move forward with the wealth of experimental data such that a complete description of energy loss in the QGP can be achieved. Measurements of jets to date clearly indicate that hard partons lose energy. Detailed comparisons of the nuclear modification factor between data and model calculations led to quantitative constraints on the opacity of the medium to hard probes. However, while there is substantial evidence for softening and broadening jets through medium interactions, the difficulties comparing measurements to theoretical calculations limit further quantitative constraints on energy loss mechanisms. Since jets are algorithmic descriptions of the initial parton, the same jet definitions must be used, including the treatment of the underlying heavy ion background, when making data and theory comparisons. An agreement is called for between theorists and experimentalists on the appropriate treatment of the background, Monte Carlo generators that enable experimental algorithms to be applied to theoretical calculations, and a clear understanding of which observables are most sensitive to the properties of the medium, even in the presence of background. This will enable us to determine the best strategy for the field to improve quantitative constraints on properties of the medium in the face of these challenges.
••
15 Oct 2018TL;DR: It is demonstrated that malicious primitive models pose immense threats to the security of ML systems, and analytical justification for the effectiveness of model-reuse attacks is provided, which points to the unprecedented complexity of today's primitive models.
Abstract: Many of today's machine learning (ML) systems are built by reusing an array of, often pre-trained, primitive models, each fulfilling distinct functionality (e.g., feature extraction). The increasing use of primitive models significantly simplifies and expedites the development cycles of ML systems. Yet, because most of such models are contributed and maintained by untrusted sources, their lack of standardization or regulation entails profound security implications, about which little is known thus far. In this paper, we demonstrate that malicious primitive models pose immense threats to the security of ML systems. We present a broad class of model-reuse attacks wherein maliciously crafted models trigger host ML systems to misbehave on targeted inputs in a highly predictable manner. By empirically studying four deep learning systems (including both individual and ensemble systems) used in skin cancer screening, speech recognition, face verification, and autonomous steering, we show that such attacks are (i) effective - the host systems misbehave on the targeted inputs as desired by the adversary with high probability, (ii) evasive - the malicious models function indistinguishably from their benign counterparts on non-targeted inputs, (iii) elastic - the malicious models remain effective regardless of various system design choices and tuning strategies, and (iv) easy - the adversary needs little prior knowledge about the data used for system tuning or inference. We provide analytical justification for the effectiveness of model-reuse attacks, which points to the unprecedented complexity of today's primitive models. This issue thus seems fundamental to many ML systems. We further discuss potential countermeasures and their challenges, which lead to several promising research directions.
••
TL;DR: In this paper, the moments of the net-kaon multiplicity distributions in relativistic heavy-ion collisions were measured and compared with Poisson and negative binomial baseline calculations as well as with UrQMD.
••
16 Sep 2018
TL;DR: A novel generative model is proposed which generates a complete radiology report automatically and is capable of not only generating high-level conclusive impressions, but also generating detailed descriptive findings sentence by sentence to support the conclusion.
Abstract: Radiologists routinely examine medical images such as X-Ray, CT, or MRI and write reports summarizing their descriptive findings and conclusive impressions. A computer-aided radiology report generation system can lighten the workload for radiologists considerably and assist them in decision making. Although the rapid development of deep learning technology makes the generation of a single conclusive sentence possible, results produced by existing methods are not sufficiently reliable due to the complexity of medical images. Furthermore, generating detailed paragraph descriptions for medical images remains a challenging problem. To tackle this problem, we propose a novel generative model which generates a complete radiology report automatically. The proposed model incorporates the Convolutional Neural Networks (CNNs) with the Long Short-Term Memory (LSTM) in a recurrent way. It is capable of not only generating high-level conclusive impressions, but also generating detailed descriptive findings sentence by sentence to support the conclusion. Furthermore, our multimodal model combines the encoding of the image and one generated sentence to construct an attention input to guide the generation of the next sentence, and henceforth maintains coherence among generated sentences. Experimental results on the publicly available Indiana U. Chest X-rays from the Open-i image collection show that our proposed recurrent attention model achieves significant improvements over baseline models according to multiple evaluation metrics.
••
University of Connecticut1, University of Exeter2, University of Oxford3, Trinity College, Dublin4, University of London5, University College London6, Arizona State University7, Adam Mickiewicz University in Poznań8, University of Cambridge9, University of California, Santa Barbara10, Lawrence University11, Santa Fe Institute12, Austrian Academy of Sciences13, Russian Academy of Sciences14, National Research University – Higher School of Economics15, Macquarie University16, Chapman University17, Harvard University18, Princeton University19, Lehigh University20, University of Texas at Austin21, Field Museum of Natural History22, University of Iceland23, National University of Singapore24, University of Pittsburgh25, University of Pennsylvania26, University of Toronto27, University of South Carolina28, Yale University29, American Museum of Natural History30
TL;DR: A database of historical and archaeological information from 30 regions around the world over the last 10,000 years revealed that characteristics, such as social scale, economy, features of governance, and information systems, show strong evolutionary relationships with each other and that complexity of a society across different world regions can be meaningfully measured using a single principal component of variation.
Abstract: Do human societies from around the world exhibit similarities in the way that they are structured, and show commonalities in the ways that they have evolved? These are long-standing questions that have proven difficult to answer. To test between competing hypotheses, we constructed a massive repository of historical and archaeological information known as “Seshat: Global History Databank.” We systematically coded data on 414 societies from 30 regions around the world spanning the last 10,000 years. We were able to capture information on 51 variables reflecting nine characteristics of human societies, such as social scale, economy, features of governance, and information systems. Our analyses revealed that these different characteristics show strong relationships with each other and that a single principal component captures around three-quarters of the observed variation. Furthermore, we found that different characteristics of social complexity are highly predictable across different world regions. These results suggest that key aspects of social organization are functionally related and do indeed coevolve in predictable ways. Our findings highlight the power of the sciences and humanities working together to rigorously test hypotheses about general rules that may have shaped human history.
••
TL;DR: A trust-region model-based algorithm for solving unconstrained stochastic optimization problems that utilizes random models of an objective function f(x), obtained from stochastically observations of the function or its gradient.
Abstract: In this paper, we propose and analyze a trust-region model-based algorithm for solving unconstrained stochastic optimization problems. Our framework utilizes random models of an objective function f(x), obtained from stochastic observations of the function or its gradient. Our method also utilizes estimates of function values to gauge progress that is being made. The convergence analysis relies on requirements that these models and these estimates are sufficiently accurate with high enough, but fixed, probability. Beyond these conditions, no assumptions are made on how these models and estimates are generated. Under these general conditions we show an almost sure global convergence of the method to a first order stationary point. In the second part of the paper, we present examples of generating sufficiently accurate random models under biased or unbiased noise assumptions. Lastly, we present some computational results showing the benefits of the proposed method compared to existing approaches that are based on sample averaging or stochastic gradients.
••
TL;DR: It is shown that in terms of the order of the accuracy, the evaluation complexity of a line-search method which is based on random first-order models and directions is the same as its counterparts that use deterministic accurate models; the use of probabilistic models only increases the complexity by a constant, which depends on the probability of the models being good.
Abstract: We present global convergence rates for a line-search method which is based on random first-order models and directions whose quality is ensured only with certain probability. We show that in terms of the order of the accuracy, the evaluation complexity of such a method is the same as its counterparts that use deterministic accurate models; the use of probabilistic models only increases the complexity by a constant, which depends on the probability of the models being good. We particularize and improve these results in the convex and strongly convex case. We also analyse a probabilistic cubic regularization variant that allows approximate probabilistic second-order models and show improved complexity bounds compared to probabilistic first-order methods; again, as a function of the accuracy, the probabilistic cubic regularization bounds are of the same (optimal) order as for the deterministic case.
•
03 Jul 2018TL;DR: In this paper, a new analysis of convergence of SGD is performed under the assumption that stochastic gradients are bounded with respect to the true gradient norm, and they also propose an alternative convergence analysis for SGD with diminishing learning rate.
Abstract: Stochastic gradient descent (SGD) is the optimization algorithm of choice in many machine learning applications such as regularized empirical risk minimization and training deep neural networks. The classical convergence analysis of SGD is carried out under the assumption that the norm of the stochastic gradient is uniformly bounded. While this might hold for some loss functions, it is always violated for cases where the objective function is strongly convex. In (Bottou et al.,2016), a new analysis of convergence of SGD is performed under the assumption that stochastic gradients are bounded with respect to the true gradient norm. Here we show that for stochastic problems arising in machine learning such bound always holds; and we also propose an alternative convergence analysis of SGD with diminishing learning rate regime, which results in more relaxed conditions than those in (Bottou et al.,2016). We then move on the asynchronous parallel setting, and prove convergence of Hogwild! algorithm in the same regime, obtaining the first convergence results for this method in the case of diminished learning rate.
••
Harvey Mudd College1, Louisiana State University2, University of Los Andes3, Queensland Museum4, James Cook University5, City University of New York6, American Museum of Natural History7, University of Giessen8, Temple University9, Florida International University10, Australian National University11, Lehigh University12, University of Macau13, University of Rhode Island14, Federal University of Ceará15
TL;DR: The results demonstrate the utility of this target‐enrichment approach to resolve phylogenetic relationships from relatively old to recent divergences and help resolve long‐standing controversial relationships in the class Anthozoa.
Abstract: Anthozoans (e.g., corals, anemones) are an ecologically important and diverse group of marine metazoans that occur from shallow to deep waters worldwide. However, our understanding of the evolutionary relationships among the similar to 7,500 species within this class is hindered by the lack of phylogenetically informative markers that can be reliably sequenced across a diversity of taxa. We designed and tested 16,306 RNA baits to capture 720 ultraconserved element loci and 1,071 exon loci. Library preparation and target enrichment were performed on 33 taxa from all orders within the class Anthozoa. Following Illumina sequencing and Trinity assembly, we recovered 1,774 of 1,791 targeted loci. The mean number of loci recovered from each species was 638 +/- 222, with more loci recovered from octocorals (783 +/- 138 loci) than hexacorals (475 +/- 187 loci). Parsimony informative sites ranged from 26 to 49% for alignments at differing hierarchical taxonomic levels (e.g., Anthozoa, Octocorallia, Hexacorallia). The per cent of variable sites within each of three genera (Acropora, Alcyonium, and Sinularia) for which multiple species were sequenced ranged from 4.7% to 30%. Maximum-likelihood analyses recovered highly resolved trees with topologies matching those supported by other studies, including the monophyly of the order Scleractinia. Our results demonstrate the utility of this target-enrichment approach to resolve phylogenetic relationships from relatively old to recent divergences. Redesigning the baits with improved affinities to capture loci within each subclass will provide a valuable toolset to address systematic questions, further our understanding of the timing of diversifications and help resolve long-standing controversial relationships in the class Anthozoa.