Abstract: Generating high-resolution, photo-realistic images has been a long-standing goal in machine learning. Recently, Nguyen et al. [37] showed one interesting way to synthesize novel images by performing gradient ascent in the latent space of a generator network to maximize the activations of one or multiple neurons in a separate classifier network. In this paper we extend this method by introducing an additional prior on the latent code, improving both sample quality and sample diversity, leading to a state-of-the-art generative model that produces high quality images at higher resolutions (227 × 227) than previous generative models, and does so for all 1000 ImageNet categories. In addition, we provide a unified probabilistic interpretation of related activation maximization methods and call the general class of models Plug and Play Generative Networks. PPGNs are composed of 1) a generator network G that is capable of drawing a wide range of image types and 2) a replaceable condition network C that tells the generator what to draw. We demonstrate the generation of images conditioned on a class (when C is an ImageNet or MIT Places classification network) and also conditioned on a caption (when C is an image captioning network). Our method also improves the state of the art of Multifaceted Feature Visualization [40], which generates the set of synthetic inputs that activate a neuron in order to better understand how deep neural networks operate. Finally, we show that our model performs reasonably well at the task of image inpainting. While image models are used in this paper, the approach is modality-agnostic and can be applied to many types of data.

Abstract: The growing explosion in the use of surveillance cameras in public security highlights the importance of vehicle search from a large-scale image or video database. However, compared with person re-identification or face recognition, vehicle search problem has long been neglected by researchers in vision community. This paper focuses on an interesting but challenging problem, vehicle re-identification (a.k.a precise vehicle search). We propose a Deep Relative Distance Learning (DRDL) method which exploits a two-branch deep convolutional network to project raw vehicle images into an Euclidean space where distance can be directly used to measure the similarity of arbitrary two vehicles. To further facilitate the future research on this problem, we also present a carefully-organized largescale image database "VehicleID", which includes multiple images of the same vehicle captured by different realworld cameras in a city. We evaluate our DRDL method on our VehicleID dataset and another recently-released vehicle model classification dataset "CompCars" in three sets of experiments: vehicle re-identification, vehicle model verification and vehicle retrieval. Experimental results show that our method can achieve promising results and outperforms several state-of-the-art approaches.

Abstract: Since its conception during the 80s, 3D-printing, also known as additive manufacturing, has been receiving unprecedented levels of attention and interest from industry and research laboratories. This is in addition to end users, who have benefited from the pervasiveness of desktop-size and relatively cheap printing machines available. 3D-printing enables almost infinite possibilities for rapid prototyping. Therefore, it has been considered for applications in numerous research fields, ranging from mechanical engineering, medicine, and materials science to chemistry. Electrochemistry is another branch of science that can certainly benefit from 3D-printing technologies, paving the way for the design and fabrication of cheaper, higher performing, and ubiquitously available electrochemical devices. Here, we aim to provide a general overview of the most commonly available 3D-printing methods along with a review of recent electrochemistry related studies adopting 3D-printing as a possible rapid prototyping fabrication tool.

Abstract: Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the cause of coronavirus disease 2019 (COVID-19), has spurred a global health crisis. To date, there are no proven options for prophylaxis for those who have been exposed to SARS-CoV-2, nor therapy for those who develop COVID-19. Immune (i.e., "convalescent") plasma refers to plasma that is collected from individuals following resolution of infection and development of antibodies. Passive antibody administration through transfusion of convalescent plasma may offer the only short-term strategy for conferring immediate immunity to susceptible individuals. There are numerous examples in which convalescent plasma has been used successfully as postexposure prophylaxis and/or treatment of infectious diseases, including other outbreaks of coronaviruses (e.g., SARS-1, Middle East respiratory syndrome [MERS]). Convalescent plasma has also been used in the COVID-19 pandemic; limited data from China suggest clinical benefit, including radiological resolution, reduction in viral loads, and improved survival. Globally, blood centers have robust infrastructure for undertaking collections and constructing inventories of convalescent plasma to meet the growing demand. Nonetheless, there are nuanced challenges, both regulatory and logistical, spanning donor eligibility, donor recruitment, collections, and transfusion itself. Data from rigorously controlled clinical trials of convalescent plasma are also few, underscoring the need to evaluate its use objectively for a range of indications (e.g., prevention vs. treatment) and patient populations (e.g., age, comorbid disease). We provide an overview of convalescent plasma, including evidence of benefit, regulatory considerations, logistical work flow, and proposed clinical trials, as scale-up is brought underway to mobilize this critical resource.

Abstract: Some cancers evade targeted therapies through a mechanism known as lineage plasticity, whereby tumor cells acquire phenotypic characteristics of a cell lineage whose survival no longer depends on the drug target. We use in vitro and in vivo human prostate cancer models to show that these tumors can develop resistance to the antiandrogen drug enzalutamide by a phenotypic shift from androgen receptor (AR)–dependent luminal epithelial cells to AR-independent basal-like cells. This lineage plasticity is enabled by the loss of TP53 and RB1 function, is mediated by increased expression of the reprogramming transcription factor SOX2, and can be reversed by restoring TP53 and RB1 function or by inhibiting SOX2 expression. Thus, mutations in tumor suppressor genes can create a state of increased cellular plasticity that, when challenged with antiandrogen therapy, promotes resistance through lineage switching.

Abstract: Pretraining large neural language models, such as BERT, has led to impressive gains on many natural language processing (NLP) tasks. However, most pretraining efforts focus on general domain corpora, such as newswire and Web. A prevailing assumption is that even domain-specific pretraining can benefit by starting from general-domain language models. In this paper, we challenge this assumption by showing that for domains with abundant unlabeled text, such as biomedicine, pretraining language models from scratch results in substantial gains over continual pretraining of general-domain language models. To facilitate this investigation, we compile a comprehensive biomedical NLP benchmark from publicly-available datasets. Our experiments show that domain-specific pretraining serves as a solid foundation for a wide range of biomedical NLP tasks, leading to new state-of-the-art results across the board. Further, in conducting a thorough evaluation of modeling choices, both for pretraining and task-specific fine-tuning, we discover that some common practices are unnecessary with BERT models, such as using complex tagging schemes in named entity recognition (NER). To help accelerate research in biomedical NLP, we have released our state-of-the-art pretrained and task-specific models for the community, and created a leaderboard featuring our BLURB benchmark (short for Biomedical Language Understanding & Reasoning Benchmark) at this https URL.

Abstract: Objective To assess the prevalence of diabetes and its risk factors. Design Population based, cross sectional study. Setting 31 provinces in mainland China with nationally representative cross sectional data from 2015 to 2017. Participants 75 880 participants aged 18 and older—a nationally representative sample of the mainland Chinese population. Main outcome measures Prevalence of diabetes among adults living in China, and the prevalence by sex, regions, and ethnic groups, estimated by the 2018 American Diabetes Association (ADA) and the World Health Organization diagnostic criteria. Demographic characteristics, lifestyle, and history of disease were recorded by participants on a questionnaire. Anthropometric and clinical assessments were made of serum concentrations of fasting plasma glucose (one measurement), two hour plasma glucose, and glycated haemoglobin (HbA1c). Results The weighted prevalence of total diabetes (n=9772), self-reported diabetes (n=4464), newly diagnosed diabetes (n=5308), and prediabetes (n=27 230) diagnosed by the ADA criteria were 12.8% (95% confidence interval 12.0% to 13.6%), 6.0% (5.4% to 6.7%), 6.8% (6.1% to 7.4%), and 35.2% (33.5% to 37.0%), respectively, among adults living in China. The weighted prevalence of total diabetes was higher among adults aged 50 and older and among men. The prevalence of total diabetes in 31 provinces ranged from 6.2% in Guizhou to 19.9% in Inner Mongolia. Han ethnicity had the highest prevalence of diabetes (12.8%) and Hui ethnicity had the lowest (6.3%) among five investigated ethnicities. The weighted prevalence of total diabetes (n=8385) using the WHO criteria was 11.2% (95% confidence interval 10.5% to 11.9%). Conclusion The prevalence of diabetes has increased slightly from 2007 to 2017 among adults living in China. The findings indicate that diabetes is an important public health problem in China.

Abstract: Given their high theoretical specific energy, lithium-oxygen batteries have received enormous attention as possible alternatives to current state-of-the-art rechargeable Li-ion batteries. However, the maximum discharge capacity in non-aqueous lithium-oxygen batteries is limited to a small fraction of its theoretical value due to the build-up of insulating lithium peroxide (Li₂O₂), the battery's primary discharge product. The discharge capacity can be increased if Li₂O₂ forms as large toroidal particles rather than as a thin conformal layer. Here, we show that trace amounts of electrolyte additives, such as H₂O, enhance the formation of Li₂O₂ toroids and result in significant improvements in capacity. Our experimental observations and a growth model show that the solvating properties of the additives prompt a solution-based mechanism that is responsible for the growth of Li₂O₂ toroids. We present a general formalism describing an additive's tendency to trigger the solution process, providing a rational design route for electrolytes that afford larger lithium-oxygen battery capacities.

Abstract: Background:The Canadian Network for Mood and Anxiety Treatments (CANMAT) conducted a revision of the 2009 guidelines by updating the evidence and recommendations. The scope of the 2016 guidelines r...

Abstract: Research on redox-flow batteries (RFBs) is currently experiencing a significant upturn, stimulated by the growing need to store increasing quantities of sustainably generated electrical energy. RFBs are promising candidates for the creation of smart grids, particularly when combined with photovoltaics and wind farms. To achieve the goal of "green", safe, and cost-efficient energy storage, research has shifted from metal-based materials to organic active materials in recent years. This Review presents an overview of various flow-battery systems. Relevant studies concerning their history are discussed as well as their development over the last few years from the classical inorganic, to organic/inorganic, to RFBs with organic redox-active cathode and anode materials. Available technologies are analyzed in terms of their technical, economic, and environmental aspects; the advantages and limitations of these systems are also discussed. Further technological challenges and prospective research possibilities are highlighted.

Abstract: Purpose: To estimate the strength and shape of the dose–response relationship between sedentary behaviour and all-cause, cardiovascular disease (CVD) and cancer mortality, and incident type 2 diabetes (T2D), adjusted for physical activity (PA). Data Sources: Pubmed, Web of Knowledge, Medline, Embase, Cochrane Library and Google Scholar (through September-2016); reference lists. Study Selection: Prospective studies reporting associations between total daily sedentary time or TV viewing time, and ≥ one outcome of interest. Data Extraction: Two independent reviewers extracted data, study quality was assessed; corresponding authors were approached where needed. Data Synthesis: Thirty-four studies (1,331,468 unique participants; good study quality) covering 8 exposure-outcome combinations were included. For total sedentary behaviour, the PA-adjusted relationship was non-linear for all-cause mortality (RR per 1 h/day: were 1.01 (1.00–1.01) ≤ 8 h/day; 1.04 (1.03–1.05) > 8 h/day of exposure), and for CVD mortality (1.01 (0.99–1.02) ≤ 6 h/day; 1.04 (1.03–1.04) > 6 h/day). The association was linear (1.01 (1.00–1.01)) with T2D and non-significant with cancer mortality. Stronger PA-adjusted associations were found for TV viewing (h/day); non-linear for all-cause mortality (1.03 (1.01–1.04) ≤ 3.5 h/day; 1.06 (1.05–1.08) > 3.5 h/day) and for CVD mortality (1.02 (0.99–1.04) ≤ 4 h/day; 1.08 (1.05–1.12) > 4 h/day). Associations with cancer mortality (1.03 (1.02–1.04)) and T2D were linear (1.09 (1.07–1.12)). Conclusions: Independent of PA, total sitting and TV viewing time are associated with greater risk for several major chronic disease outcomes. For all-cause and CVD mortality, a threshold of 6–8 h/day of total sitting and 3–4 h/day of TV viewing was identified, above which the risk is increased.

Abstract: The accurate evaluation of crash causal factors can provide fundamental information for effective transportation policy, vehicle design, and driver education. Naturalistic driving (ND) data collected with multiple onboard video cameras and sensors provide a unique opportunity to evaluate risk factors during the seconds leading up to a crash. This paper uses a National Academy of Sciences-sponsored ND dataset comprising 905 injurious and property damage crash events, the magnitude of which allows the first direct analysis (to our knowledge) of causal factors using crashes only. The results show that crash causation has shifted dramatically in recent years, with driver-related factors (i.e., error, impairment, fatigue, and distraction) present in almost 90% of crashes. The results also definitively show that distraction is detrimental to driver safety, with handheld electronic devices having high use rates and risk.

Abstract: The highest two-qubit gate fidelities have been demonstrated in two experiments that use scalable trapped ion platforms.

Abstract: Phosphorene, an emerging two-dimensional material, has received considerable attention due to its layer-controlled direct bandgap, high carrier mobility, negative Poisson's ratio and unique in-plane anisotropy. As cousins of phosphorene, 2D group-VA arsenene, antimonene and bismuthene have also garnered tremendous interest due to their intriguing structures and fascinating electronic properties. 2D group-VA family members are opening up brand-new opportunities for their multifunctional applications encompassing electronics, optoelectronics, topological spintronics, thermoelectrics, sensors, Li- or Na-batteries. In this review, we extensively explore the latest theoretical and experimental progress made in the fundamental properties, fabrications and applications of 2D group-VA materials, and offer perspectives and challenges for the future of this emerging field.

Abstract: Masks and testing are necessary to combat asymptomatic spread in aerosols and droplets Respiratory infections occur through the transmission of virus-containing droplets (>5 to 10 µm) and aerosols (≤5 µm) exhaled from infected individuals during breathing, speaking, coughing, and sneezing. Traditional respiratory disease control measures are designed to reduce transmission by droplets produced in the sneezes and coughs of infected individuals. However, a large proportion of the spread of coronavirus disease 2019 (COVID-19) appears to be occurring through airborne transmission of aerosols produced by asymptomatic individuals during breathing and speaking (1—3). Aerosols can accumulate, remain infectious in indoor air for hours, and be easily inhaled deep into the lungs. For society to resume, measures designed to reduce aerosol transmission must be implemented, including universal masking and regular, widespread testing to identify and isolate infected asymptomatic individuals.

Abstract: Saliency methods have emerged as a popular tool to highlight features in an input deemed relevant for the prediction of a learned model. Several saliency methods have been proposed, often guided by visual appeal on image data. In this work, we propose an actionable methodology to evaluate what kinds of explanations a given method can and cannot provide. We find that reliance, solely, on visual assessment can be misleading. Through extensive experiments we show that some existing saliency methods are independent both of the model and of the data generating process. Consequently, methods that fail the proposed tests are inadequate for tasks that are sensitive to either data or model, such as, finding outliers in the data, explaining the relationship between inputs and outputs that the model learned, and debugging the model. We interpret our findings through an analogy with edge detection in images, a technique that requires neither training data nor model. Theory in the case of a linear model and a single-layer convolutional neural network supports our experimental findings.

Abstract: Molecular dynamics (MD) simulation is a valuable tool for characterizing the structural dynamics of folded proteins and should be similarly applicable to disordered proteins and proteins with both folded and disordered regions. It has been unclear, however, whether any physical model (force field) used in MD simulations accurately describes both folded and disordered proteins. Here, we select a benchmark set of 21 systems, including folded and disordered proteins, simulate these systems with six state-of-the-art force fields, and compare the results to over 9,000 available experimental data points. We find that none of the tested force fields simultaneously provided accurate descriptions of folded proteins, of the dimensions of disordered proteins, and of the secondary structure propensities of disordered proteins. Guided by simulation results on a subset of our benchmark, however, we modified parameters of one force field, achieving excellent agreement with experiment for disordered proteins, while maintaining state-of-the-art accuracy for folded proteins. The resulting force field, a99SB-disp, should thus greatly expand the range of biological systems amenable to MD simulation. A similar approach could be taken to improve other force fields.

Abstract: Mobile-edge computation offloading (MECO) offloads intensive mobile computation to clouds located at the edges of cellular networks. Thereby, MECO is envisioned as a promising technique for prolonging the battery lives and enhancing the computation capacities of mobiles. In this paper, we study resource allocation for a multiuser MECO system based on time-division multiple access (TDMA) and orthogonal frequency-division multiple access (OFDMA). First, for the TDMA MECO system with infinite or finite computation capacity, the optimal resource allocation is formulated as a convex optimization problem for minimizing the weighted sum mobile energy consumption under the constraint on computation latency. The optimal policy is proved to have a threshold-based structure with respect to a derived offloading priority function, which yields priorities for users according to their channel gains and local computing energy consumption. As a result, users with priorities above and below a given threshold perform complete and minimum offloading, respectively. Moreover, for the cloud with finite capacity, a sub-optimal resource-allocation algorithm is proposed to reduce the computation complexity for computing the threshold. Next, we consider the OFDMA MECO system, for which the optimal resource allocation is formulated as a non-convex mixed-integer problem. To solve this challenging problem and characterize its policy structure, a sub-optimal low-complexity algorithm is proposed by transforming the OFDMA problem to its TDMA counterpart. The corresponding resource allocation is derived by defining an average offloading priority function and shown to have close-to-optimal performance by simulation.

Abstract: We present two approaches to use unlabeled data to improve Sequence Learning with recurrent networks. The first approach is to predict what comes next in a sequence, which is a language model in NLP. The second approach is to use a sequence autoencoder, which reads the input sequence into a vector and predicts the input sequence again. These two algorithms can be used as a "pretraining" algorithm for a later supervised sequence learning algorithm. In other words, the parameters obtained from the pretraining step can then be used as a starting point for other supervised training models. In our experiments, we find that long short term memory recurrent networks after pretrained with the two approaches become more stable to train and generalize better. With pretraining, we were able to achieve strong performance in many classification tasks, such as text classification with IMDB, DBpedia or image recognition in CIFAR-10.

Abstract: The functions of epithelial tissues are dictated by the types, abundance and distribution of the differentiated cells they contain. Attempts to restore tissue function after damage require knowledge of how physiological tasks are distributed among cell types, and how cell states vary between homeostasis, injury–repair and disease. In the conducting airway, a heterogeneous basal cell population gives rise to specialized luminal cells that perform mucociliary clearance1. Here we perform single-cell profiling of human bronchial epithelial cells and mouse tracheal epithelial cells to obtain a comprehensive census of cell types in the conducting airway and their behaviour in homeostasis and regeneration. Our analysis reveals cell states that represent known and novel cell populations, delineates their heterogeneity and identifies distinct differentiation trajectories during homeostasis and tissue repair. Finally, we identified a novel, rare cell type that we call the ‘pulmonary ionocyte’, which co-expresses FOXI1, multiple subunits of the vacuolar-type H+-ATPase (V-ATPase) and CFTR, the gene that is mutated in cystic fibrosis. Using immunofluorescence, modulation of signalling pathways and electrophysiology, we show that Notch signalling is necessary and FOXI1 expression is sufficient to drive the production of the pulmonary ionocyte, and that the pulmonary ionocyte is a major source of CFTR activity in the conducting airway epithelium. Single-cell RNA sequencing analysis is used to identify cell types in the tracheal epithelium, including previously unidentified ionocytes, which express high levels of the cystic fibrosis transmembrane conductance regulator, CFTR.

Abstract: Scaling up deep neural network capacity has been known as an effective approach to improving model quality for several different machine learning tasks. In many cases, increasing model capacity beyond the memory limit of a single accelerator has required developing special algorithms or infrastructure. These solutions are often architecture-specific and do not transfer to other tasks. To address the need for efficient and task-independent model parallelism, we introduce GPipe, a pipeline parallelism library that allows scaling any network that can be expressed as a sequence of layers. By pipelining different sub-sequences of layers on separate accelerators, GPipe provides the flexibility of scaling a variety of different networks to gigantic sizes efficiently. Moreover, GPipe utilizes a novel batch-splitting pipelining algorithm, resulting in almost linear speedup when a model is partitioned across multiple accelerators. We demonstrate the advantages of GPipe by training large-scale neural networks on two different tasks with distinct network architectures: (i) Image Classification: We train a 557-million-parameter AmoebaNet model and attain a top-1 accuracy of 84.4% on ImageNet-2012, (ii) Multilingual Neural Machine Translation: We train a single 6-billion-parameter, 128-layer Transformer model on a corpus spanning over 100 languages and achieve better quality than all bilingual models.

Abstract: Visual understanding goes well beyond object recognition. With one glance at an image, we can effortlessly imagine the world beyond the pixels: for instance, we can infer people's actions, goals, and mental states. While this task is easy for humans, it is tremendously difficult for today's vision systems, requiring higher-order cognition and commonsense reasoning about the world. We formalize this task as Visual Commonsense Reasoning. Given a challenging question about an image, a machine must answer correctly and then provide a rationale justifying its answer. Next, we introduce a new dataset, VCR, consisting of 290k multiple choice QA problems derived from 110k movie scenes. The key recipe for generating non-trivial and high-quality problems at scale is Adversarial Matching, a new approach to transform rich annotations into multiple choice questions with minimal bias. Experimental results show that while humans find VCR easy (over 90% accuracy), state-of-the-art vision models struggle (~45%). To move towards cognition-level understanding, we present a new reasoning engine, Recognition to Cognition Networks (R2C), that models the necessary layered inferences for grounding, contextualization, and reasoning. R2C helps narrow the gap between humans and machines (~65%); still, the challenge is far from solved, and we provide analysis that suggests avenues for future work.

Abstract: T-cell exhaustion was originally identified during chronic infection in mice, and was subsequently observed in humans with cancer. The exhausted T cells in the tumor microenvironment show overexpressed inhibitory receptors, decreased effector cytokine production and cytolytic activity, leading to the failure of cancer elimination. Restoring exhausted T cells represents an inspiring strategy for cancer treatment, which has yielded promising results and become a significant breakthrough in the cancer immunotherapy. In this review, we overview the updated understanding on the exhausted T cells in cancer and their potential regulatory mechanisms and discuss current therapeutic interventions targeting exhausted T cells in clinical trials.