Abstract: Self-supervised learning has become increasingly important to leverage the abundance of unlabeled data available on platforms like YouTube. Whereas most existing approaches learn low-level representations, we propose a joint visual-linguistic model to learn high-level features without any explicit supervision. In particular, inspired by its recent success in language modeling, we build upon the BERT model to learn bidirectional joint distributions over sequences of visual and linguistic tokens, derived from vector quantization of video data and off-the-shelf speech recognition outputs, respectively. We use VideoBERT in numerous tasks, including action classification and video captioning. We show that it can be applied directly to open-vocabulary classification, and confirm that large amounts of training data and cross-modal information are critical to performance. Furthermore, we outperform the state-of-the-art on video captioning, and quantitative results verify that the model learns high-level semantic features.

Abstract: Background A point-prevalence survey that was conducted in the United States in 2011 showed that 4% of hospitalized patients had a health care–associated infection. We repeated the survey ...

Abstract: The FANTOM5 project investigates transcription initiation activities in more than 1,000 human and mouse primary cells, cell lines and tissues using CAGE. Based on manual curation of sample information and development of an ontology for sample classification, we assemble the resulting data into a centralized data resource (http://fantom.gsc.riken.jp/5/). This resource contains web-based tools and data-access points for the research community to search and extract data related to samples, genes, promoter activities, transcription factors and enhancers across the FANTOM5 atlas.

Abstract: Two-dimensional semiconductors such as MoS2 are an emerging material family with wide-ranging potential applications in electronics, optoelectronics, and energy harvesting. Large-area growth methods are needed to open the way to applications. Control over lattice orientation during growth remains a challenge. This is needed to minimize or even avoid the formation of grain boundaries, detrimental to electrical, optical, and mechanical properties of MoS2 and other 2D semiconductors. Here, we report on the growth of high-quality monolayer MoS2 with control over lattice orientation. We show that the monolayer film is composed of coalescing single islands with limited numbers of lattice orientation due to an epitaxial growth mechanism. Optical absorbance spectra acquired over large areas show significant absorbance in the high-energy part of the spectrum, indicating that MoS2 could also be interesting for harvesting this region of the solar spectrum and fabrication of UV-sensitive photodetectors. Even though t...

Abstract: We show that the bulk-boundary correspondence for topological insulators can be modified in the presence of non-Hermiticity. We consider a one-dimensional tight-binding model with gain and loss as well as long-range hopping. The system is described by a non-Hermitian Hamiltonian that encircles an exceptional point in momentum space. The winding number has a fractional value of 1/2. There is only one dynamically stable zero-energy edge state due to the defectiveness of the Hamiltonian. This edge state is robust to disorder due to protection by a chiral symmetry. We also discuss experimental realization with arrays of coupled resonator optical waveguides.

Abstract: Deep neural networks and machine-learning algorithms are pervasively used in several applications, ranging from computer vision to computer security. In most of these applications, the learning algorithm has to face intelligent and adaptive attackers who can carefully manipulate data to purposely subvert the learning process. As these algorithms have not been originally designed under such premises, they have been shown to be vulnerable to well-crafted, sophisticated attacks, including training-time poisoning and test-time evasion attacks (also known as adversarial examples). The problem of countering these threats and learning secure classifiers in adversarial settings has thus become the subject of an emerging, relevant research field known as adversarial machine learning. The purposes of this tutorial are: (a) to introduce the fundamentals of adversarial machine learning to the security community; (b) to illustrate the design cycle of a learning-based pattern recognition system for adversarial tasks; (c) to present novel techniques that have been recently proposed to assess performance of pattern classifiers and deep learning algorithms under attack, evaluate their vulnerabilities, and implement defense strategies that make learning algorithms more robust to attacks; and (d) to show some applications of adversarial machine learning to pattern recognition tasks like object recognition in images, biometric identity recognition, spam and malware detection.

Abstract: Learning long term dependencies in recurrent networks is difficult due to vanishing and exploding gradients. To overcome this difficulty, researchers have developed sophisticated optimization techniques and network architectures. In this paper, we propose a simpler solution that use recurrent neural networks composed of rectified linear units. Key to our solution is the use of the identity matrix or its scaled version to initialize the recurrent weight matrix. We find that our solution is comparable to LSTM on our four benchmarks: two toy problems involving long-range temporal structures, a large language modeling problem and a benchmark speech recognition problem.

Abstract: Simultaneously accurate and efficient prediction of molecular properties throughout chemical compound space is a critical ingredient toward rational compound design in chemical and pharmaceutical industries. Aiming toward this goal, we develop and apply a systematic hierarchy of efficient empirical methods to estimate atomization and total energies of molecules. These methods range from a simple sum over atoms, to addition of bond energies, to pairwise interatomic force fields, reaching to the more sophisticated machine learning approaches that are capable of describing collective interactions between many atoms or bonds. In the case of equilibrium molecular geometries, even simple pairwise force fields demonstrate prediction accuracy comparable to benchmark energies calculated using density functional theory with hybrid exchange-correlation functionals; however, accounting for the collective many-body interactions proves to be essential for approaching the “holy grail” of chemical accuracy of 1 kcal/mol ...

Abstract: Much of our understanding of Earth's past climate comes from the measurement of oxygen and carbon isotope variations in deep-sea benthic foraminifera. Yet, long intervals in existing records lack the temporal resolution and age control needed to thoroughly categorize climate states of the Cenozoic era and to study their dynamics. Here, we present a new, highly resolved, astronomically dated, continuous composite of benthic foraminifer isotope records developed in our laboratories. Four climate states-Hothouse, Warmhouse, Coolhouse, Icehouse-are identified on the basis of their distinctive response to astronomical forcing depending on greenhouse gas concentrations and polar ice sheet volume. Statistical analysis of the nonlinear behavior encoded in our record reveals the key role that polar ice volume plays in the predictability of Cenozoic climate dynamics.

Abstract: BackgroundIn light of the worldwide increase in childhood obesity, we examined the association between body-mass index (BMI) in late adolescence and death from cardiovascular causes in adulthood. MethodsWe grouped data on BMI, as measured from 1967 through 2010 in 2.3 million Israeli adolescents (mean age, 17.3±0.4 years), according to age- and sex-specific percentiles from the U.S. Centers for Disease Control and Prevention. Primary outcomes were the number of deaths attributed to coronary heart disease, stroke, sudden death from an unknown cause, or a combination of all three categories (total cardiovascular causes) by mid-2011. Cox proportional-hazards models were used. ResultsDuring 42,297,007 person-years of follow-up, 2918 of 32,127 deaths (9.1%) were from cardiovascular causes, including 1497 from coronary heart disease, 528 from stroke, and 893 from sudden death. On multivariable analysis, there was a graded increase in the risk of death from cardiovascular causes and all causes that started among...

Abstract: Importance Establishment of the infant microbiome has lifelong implications on health and immunity. Gut microbiota of breastfed compared with nonbreastfed individuals differ during infancy as well as into adulthood. Breast milk contains a diverse population of bacteria, but little is known about the vertical transfer of bacteria from mother to infant by breastfeeding. Objective To determine the association between the maternal breast milk and areolar skin and infant gut bacterial communities. Design, Setting, and Participants In a prospective, longitudinal study, bacterial composition was identified with sequencing of the 16S ribosomal RNA gene in breast milk, areolar skin, and infant stool samples of 107 healthy mother-infant pairs. The study was conducted in Los Angeles, California, and St Petersburg, Florida, between January 1, 2010, and February 28, 2015. Exposures Amount and duration of daily breastfeeding and timing of solid food introduction. Main Outcomes and Measures Bacterial composition in maternal breast milk, areolar skin, and infant stool by sequencing of the 16S ribosomal RNA gene. Results In the 107 healthy mother and infant pairs (median age at the time of specimen collection, 40 days; range, 1-331 days), 52 (43.0%) of the infants were male. Bacterial communities were distinct in milk, areolar skin, and stool, differing in both composition and diversity. The infant gut microbial communities were more closely related to an infant’s mother’s milk and skin compared with a random mother (mean difference in Bray-Curtis distances, 0.012 and 0.014, respectively; P P = .003) and composition changes were associated with the proportion of daily breast milk intake in a dose-dependent manner, even after the introduction of solid foods. Conclusions and Relevance The results of this study indicate that bacteria in mother’s breast milk seed the infant gut, underscoring the importance of breastfeeding in the development of the infant gut microbiome.

Abstract: Insect pests substantially reduce yields of three staple grains—rice, maize, and wheat—but models assessing the agricultural impacts of global warming rarely consider crop losses to insects. We use established relationships between temperature and the population growth and metabolic rates of insects to estimate how and where climate warming will augment losses of rice, maize, and wheat to insects. Global yield losses of these grains are projected to increase by 10 to 25% per degree of global mean surface warming. Crop losses will be most acute in areas where warming increases both population growth and metabolic rates of insects. These conditions are centered primarily in temperate regions, where most grain is produced.

Abstract: The low toxicity and a near-ideal choice of bandgap make tin perovskite an attractive alternative to lead perovskite in low cost solar cells. However, the development of Sn perovskite solar cells has been impeded by their extremely poor stability when exposed to oxygen. We report low-dimensional Sn perovskites that exhibit markedly enhanced air stability in comparison with their 3D counterparts. The reduced degradation under air exposure is attributed to the improved thermodynamic stability after dimensional reduction, the encapsulating organic ligands, and the compact perovskite film preventing oxygen ingress. We then explore these highly oriented low-dimensional Sn perovskite films in solar cells. The perpendicular growth of the perovskite domains between electrodes allows efficient charge carrier transport, leading to power conversion efficiencies of 5.94% without the requirement of further device structure engineering. We tracked the performance of unencapsulated devices over 100 h and found no apprec...

Abstract: We provide a current perspective on the rapidly developing field of Majorana zero modes in solid state systems. We emphasize the theoretical prediction, experimental realization, and potential use of Majorana zero modes in future information processing devices through braiding-based topological quantum computation. Well-separated Majorana zero modes should manifest non-Abelian braiding statistics suitable for unitary gate operations for topological quantum computation. Recent experimental work, following earlier theoretical predictions, has shown specific signatures consistent with the existence of Majorana modes localized at the ends of semiconductor nanowires in the presence of superconducting proximity effect. We discuss the experimental findings and their theoretical analyses, and provide a perspective on the extent to which the observations indicate the existence of anyonic Majorana zero modes in solid state systems. We also discuss fractional quantum Hall systems (the 5/2 state) in this context. We describe proposed schemes for carrying out braiding with Majorana zero modes as well as the necessary steps for implementing topological quantum computation.

Abstract: A minimal cell can be thought of as comprising informational, compartment-forming and metabolic subsystems. To imagine the abiotic assembly of such an overall system, however, places great demands on hypothetical prebiotic chemistry. The perceived differences and incompatibilities between these subsystems have led to the widely held assumption that one or other subsystem must have preceded the others. Here we experimentally investigate the validity of this assumption by examining the assembly of various biomolecular building blocks from prebiotically plausible intermediates and one-carbon feedstock molecules. We show that precursors of ribonucleotides, amino acids and lipids can all be derived by the reductive homologation of hydrogen cyanide and some of its derivatives, and thus that all the cellular subsystems could have arisen simultaneously through common chemistry. The key reaction steps are driven by ultraviolet light, use hydrogen sulfide as the reductant and can be accelerated by Cu(I)–Cu(II) photoredox cycling. A minimal cell — one that has all the minimum requirements for life — is still a complex entity comprising informational, compartment-forming and metabolic subsystems. Here it is shown that, contrary to previous assumptions, a common prebiotically plausible chemistry can give rise to building blocks for all the subsystems.

Abstract: Semi-supervised learning has proven to be a powerful paradigm for leveraging unlabeled data to mitigate the reliance on large labeled datasets. In this work, we unify the current dominant approaches for semi-supervised learning to produce a new algorithm, MixMatch, that guesses low-entropy labels for data-augmented unlabeled examples and mixes labeled and unlabeled data using MixUp. MixMatch obtains state-of-the-art results by a large margin across many datasets and labeled data amounts. For example, on CIFAR-10 with 250 labels, we reduce error rate by a factor of 4 (from 38% to 11%) and by a factor of 2 on STL-10. We also demonstrate how MixMatch can help achieve a dramatically better accuracy-privacy trade-off for differential privacy. Finally, we perform an ablation study to tease apart which components of MixMatch are most important for its success. Code is attached.

Abstract: In open-domain dialogue intelligent agents should exhibit the use of knowledge, however there are few convincing demonstrations of this to date. The most popular sequence to sequence models typically "generate and hope" generic utterances that can be memorized in the weights of the model when mapping from input utterance(s) to output, rather than employing recalled knowledge as context. Use of knowledge has so far proved difficult, in part because of the lack of a supervised learning benchmark task which exhibits knowledgeable open dialogue with clear grounding. To that end we collect and release a large dataset with conversations directly grounded with knowledge retrieved from Wikipedia. We then design architectures capable of retrieving knowledge, reading and conditioning on it, and finally generating natural responses. Our best performing dialogue models are able to conduct knowledgeable discussions on open-domain topics as evaluated by automatic metrics and human evaluations, while our new benchmark allows for measuring further improvements in this important research direction.

Abstract: Abstract

Abstract: Clearing senescent cells extends healthspan in mice. Using a hypothesis-driven bioinformatics-based approach, we recently identified pro-survival pathways in human senescent cells that contribute to their resistance to apoptosis. This led to identification of dasatinib (D) and quercetin (Q) as senolytics, agents that target some of these pathways and induce apoptosis preferentially in senescent cells. Among other pro-survival regulators identified was Bcl-xl. Here, we tested whether the Bcl-2 family inhibitors, navitoclax (N) and TW-37 (T), are senolytic. Like D and Q, N is senolytic in some, but not all types of senescent cells: N reduced viability of senescent human umbilical vein epithelial cells (HUVECs), IMR90 human lung fibroblasts, and murine embryonic fibroblasts (MEFs), but not human primary preadipocytes, consistent with our previous finding that Bcl-xl siRNA is senolytic in HUVECs, but not preadipocytes. In contrast, T had little senolytic activity. N targets Bcl-2, Bcl-xl, and Bcl-w, while T targets Bcl-2, Bcl-xl, and Mcl-1. The combination of Bcl-2, Bcl-xl, and Bcl-w siRNAs was senolytic in HUVECs and IMR90 cells, while combination of Bcl-2, Bcl-xl, and Mcl-1 siRNAs was not. Susceptibility to N correlated with patterns of Bcl-2 family member proteins in different types of human senescent cells, as has been found in predicting response of cancers to N. Thus, N is senolytic and acts in a potentially predictable cell type-restricted manner. The hypothesis-driven, bioinformatics-based approach we used to discover that dasatinib (D) and quercetin (Q) are senolytic can be extended to increase the repertoire of senolytic drugs, including additional cell type-specific senolytic agents.

Abstract: The development of advanced flexible large-area electronics such as flexible displays and sensors will thrive on engineered functional ink formulations for printed electronics where the spontaneous arrangement of molecules aids the printing processes. Here we report a printable elastic conductor with a high initial conductivity of 738 S cm−1 and a record high conductivity of 182 S cm−1 when stretched to 215% strain. The elastic conductor ink is comprised of Ag flakes, a fluorine rubber and a fluorine surfactant. The fluorine surfactant constitutes a key component which directs the formation of surface-localized conductive networks in the printed elastic conductor, leading to a high conductivity and stretchability. We demonstrate the feasibility of our inks by fabricating a stretchable organic transistor active matrix on a rubbery stretchability-gradient substrate with unimpaired functionality when stretched to 110%, and a wearable electromyogram sensor printed onto a textile garment. Printable electronics is highly desirable for high throughput device manufacture. Here, Matsuhisa et al. report an electric ink, made of a self-assembled network of sliver flakes on the surface of a fluorine rubber matrix, which exhibits high conductivity and mechanical durability to achieve this goal.

Abstract: The application of nanotechnology for the treatment of cancer is mostly based on early tumor detection and diagnosis by nanodevices capable of selective targeting and delivery of chemotherapeutic drugs to the specific tumor site Due to the remarkable properties of gold nanoparticles, they have long been considered as a potential tool for diagnosis of various cancers and for drug delivery applications These properties include high surface area to volume ratio, surface plasmon resonance, surface chemistry and multi-functionalization, facile synthesis, and stable nature Moreover, the non-toxic and non-immunogenic nature of gold nanoparticles and the high permeability and retention effect provide additional benefits by enabling easy penetration and accumulation of drugs at the tumor sites Various innovative approaches with gold nanoparticles are under development In this review, we provide an overview of recent progress made in the application of gold nanoparticles in the treatment of cancer by tumor detection, drug delivery, imaging, photothermal and photodynamic therapy and their current limitations in terms of bioavailability and the fate of the nanoparticles

Abstract: The consolidation of the scientific publishing industry has been the topic of much debate within and outside the scientific community, especially in relation to major publishers’ high profit margins. However, the share of scientific output published in the journals of these major publishers, as well as its evolution over time and across various disciplines, has not yet been analyzed. This paper provides such analysis, based on 45 million documents indexed in the Web of Science over the period 1973-2013. It shows that in both natural and medical sciences (NMS) and social sciences and humanities (SSH), Reed-Elsevier, Wiley-Blackwell, Springer, and Taylor & Francis increased their share of the published output, especially since the advent of the digital era (mid-1990s). Combined, the top five most prolific publishers account for more than 50% of all papers published in 2013. Disciplines of the social sciences have the highest level of concentration (70% of papers from the top five publishers), while the humanities have remained relatively independent (20% from top five publishers). NMS disciplines are in between, mainly because of the strength of their scientific societies, such as the ACS in chemistry or APS in physics. The paper also examines the migration of journals between small and big publishing houses and explores the effect of publisher change on citation impact. It concludes with a discussion on the economics of scholarly publishing.

Abstract: Importance Sepsis is a heterogeneous syndrome. Identification of distinct clinical phenotypes may allow more precise therapy and improve care. Objective To derive sepsis phenotypes from clinical data, determine their reproducibility and correlation with host-response biomarkers and clinical outcomes, and assess the potential causal relationship with results from randomized clinical trials (RCTs). Design, Settings, and Participants Retrospective analysis of data sets using statistical, machine learning, and simulation tools. Phenotypes were derived among 20 189 total patients (16 552 unique patients) who met Sepsis-3 criteria within 6 hours of hospital presentation at 12 Pennsylvania hospitals (2010-2012) using consensuskmeans clustering applied to 29 variables. Reproducibility and correlation with biological parameters and clinical outcomes were assessed in a second database (2013-2014; n = 43 086 total patients and n = 31 160 unique patients), in a prospective cohort study of sepsis due to pneumonia (n = 583), and in 3 sepsis RCTs (n = 4737). Exposures All clinical and laboratory variables in the electronic health record. Main Outcomes and Measures Derived phenotype (α, β, γ,and δ) frequency, host-response biomarkers, 28-day and 365-day mortality, and RCT simulation outputs. Results The derivation cohort included 20 189 patients with sepsis (mean age, 64 [SD, 17] years; 10 022 [50%] male; mean maximum 24-hour Sequential Organ Failure Assessment [SOFA] score, 3.9 [SD, 2.4]). The validation cohort included 43 086 patients (mean age, 67 [SD, 17] years; 21 993 [51%] male; mean maximum 24-hour SOFA score, 3.6 [SD, 2.0]). Of the 4 derived phenotypes, the α phenotype was the most common (n = 6625; 33%) and included patients with the lowest administration of a vasopressor; in the β phenotype (n = 5512; 27%), patients were older and had more chronic illness and renal dysfunction; in the γ phenotype (n = 5385; 27%), patients had more inflammation and pulmonary dysfunction; and in the δ phenotype (n = 2667; 13%), patients had more liver dysfunction and septic shock. Phenotype distributions were similar in the validation cohort. There were consistent differences in biomarker patterns by phenotype. In the derivation cohort, cumulative 28-day mortality was 287 deaths of 5691 unique patients (5%) for the α phenotype; 561 of 4420 (13%) for the β phenotype; 1031 of 4318 (24%) for the γ phenotype; and 897 of 2223 (40%) for the δ phenotype. Across all cohorts and trials, 28-day and 365-day mortality were highest among the δ phenotype vs the other 3 phenotypes (P 33% chance of benefit to >60% chance of harm). Conclusions and Relevance In this retrospective analysis of data sets from patients with sepsis, 4 clinical phenotypes were identified that correlated with host-response patterns and clinical outcomes, and simulations suggested these phenotypes may help in understanding heterogeneity of treatment effects. Further research is needed to determine the utility of these phenotypes in clinical care and for informing trial design and interpretation.

Abstract: Recently, polymerization-induced self-assembly (PISA) has become widely recognized as a robust and efficient route to produce block copolymer nanoparticles of controlled size, morphology, and surface chemistry. Several reviews of this field have been published since 2012, but a substantial number of new papers have been published in the last three years. In this Perspective, we provide a critical appraisal of the various advantages offered by this approach, while also pointing out some of its current drawbacks. Promising future research directions as well as remaining technical challenges and unresolved problems are briefly highlighted.