Abstract: Adeno-associated virus (AAV) vectors are the leading platform for gene delivery for the treatment of a variety of human diseases. Recent advances in developing clinically desirable AAV capsids, optimizing genome designs and harnessing revolutionary biotechnologies have contributed substantially to the growth of the gene therapy field. Preclinical and clinical successes in AAV-mediated gene replacement, gene silencing and gene editing have helped AAV gain popularity as the ideal therapeutic vector, with two AAV-based therapeutics gaining regulatory approval in Europe or the United States. Continued study of AAV biology and increased understanding of the associated therapeutic challenges and limitations will build the foundation for future clinical success.

Abstract: In de novo drug design, computational strategies are used to generate novel molecules with good affinity to the desired biological target. In this work, we show that recurrent neural networks can be trained as generative models for molecular structures, similar to statistical language models in natural language processing. We demonstrate that the properties of the generated molecules correlate very well with the properties of the molecules used to train the model. In order to enrich libraries with molecules active toward a given biological target, we propose to fine-tune the model with small sets of molecules, which are known to be active against that target. Against Staphylococcus aureus, the model reproduced 14% of 6051 hold-out test molecules that medicinal chemists designed, whereas against Plasmodium falciparum (Malaria), it reproduced 28% of 1240 test molecules. When coupled with a scoring function, our model can perform the complete de novo drug design cycle to generate large sets of novel molecule...

Abstract: In this work, we present a new network design paradigm. Our goal is to help advance the understanding of network design and discover design principles that generalize across settings. Instead of focusing on designing individual network instances, we design network design spaces that parametrize populations of networks. The overall process is analogous to classic manual design of networks, but elevated to the design space level. Using our methodology we explore the structure aspect of network design and arrive at a low-dimensional design space consisting of simple, regular networks that we call RegNet. The core insight of the RegNet parametrization is surprisingly simple: widths and depths of good networks can be explained by a quantized linear function. We analyze the RegNet design space and arrive at interesting findings that do not match the current practice of network design. The RegNet design space provides simple and fast networks that work well across a wide range of flop regimes. Under comparable training settings and flops, the RegNet models outperform the popular EfficientNet models while being up to 5x faster on GPUs.

Abstract: We introduce the "Energy-based Generative Adversarial Network" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce contrastive samples with minimal energies, while the discriminator is trained to assign high energies to these generated samples. Viewing the discriminator as an energy function allows to use a wide variety of architectures and loss functionals in addition to the usual binary classifier with logistic output. Among them, we show one instantiation of EBGAN framework as using an auto-encoder architecture, with the energy being the reconstruction error, in place of the discriminator. We show that this form of EBGAN exhibits more stable behavior than regular GANs during training. We also show that a single-scale architecture can be trained to generate high-resolution images.

Abstract: Recent progress on salient object detection is substantial, benefiting mostly from the explosive development of Convolutional Neural Networks (CNNs). Semantic segmentation and salient object detection algorithms developed lately have been mostly based on Fully Convolutional Neural Networks (FCNs). There is still a large room for improvement over the generic FCN models that do not explicitly deal with the scale-space problem. The Holistically-Nested Edge Detector (HED) provides a skip-layer structure with deep supervision for edge and boundary detection, but the performance gain of HED on saliency detection is not obvious. In this paper, we propose a new salient object detection method by introducing short connections to the skip-layer structures within the HED architecture. Our framework takes full advantage of multi-level and multi-scale features extracted from FCNs, providing more advanced representations at each layer, a property that is critically needed to perform segment detection. Our method produces state-of-the-art results on 5 widely tested salient object detection benchmarks, with advantages in terms of efficiency (0.08 seconds per image), effectiveness, and simplicity over the existing algorithms. Beyond that, we conduct an exhaustive analysis of the role of training data on performance. We provide a training set for future research and fair comparisons.

Abstract: Background Anti-CD19 chimeric antigen receptor (CAR) T-cell therapy has shown remarkable clinical efficacy in B-cell cancers. However, CAR T cells can induce substantial toxic effects, and...

Abstract: Summary Background Abiraterone acetate plus prednisone significantly improved radiographic progression-free survival compared with placebo plus prednisone in men with chemotherapy-naive castration-resistant prostate cancer at the interim analyses of the COU-AA-302 trial. Here, we present the prespecified final analysis of the trial, assessing the effect of abiraterone acetate plus prednisone on overall survival, time to opiate use, and use of other subsequent therapies. Methods In this placebo-controlled, double-blind, randomised phase 3 study, 1088 asymptomatic or mildly symptomatic patients with chemotherapy-naive prostate cancer stratified by Eastern Cooperative Oncology performance status (0 vs 1) were randomly assigned with a permuted block allocation scheme via a web response system in a 1:1 ratio to receive either abiraterone acetate (1000 mg once daily) plus prednisone (5 mg twice daily; abiraterone acetate group) or placebo plus prednisone (placebo group). Coprimary endpoints were radiographic progression-free survival and overall survival analysed in the intention-to-treat population. The study is registered with ClinicalTrials.gov, number NCT00887198. Findings At a median follow-up of 49·2 months (IQR 47·0–51·8), 741 (96%) of the prespecified 773 death events for the final analysis had been observed: 354 (65%) of 546 patients in the abiraterone acetate group and 387 (71%) of 542 in the placebo group. 238 (44%) patients initially receiving prednisone alone subsequently received abiraterone acetate plus prednisone as crossover per protocol (93 patients) or as subsequent therapy (145 patients). Overall, 365 (67%) patients in the abiraterone acetate group and 435 (80%) in the placebo group received subsequent treatment with one or more approved agents. Median overall survival was significantly longer in the abiraterone acetate group than in the placebo group (34·7 months [95% CI 32·7–36·8] vs 30·3 months [28·7–33·3]; hazard ratio 0·81 [95% CI 0·70–0·93]; p=0·0033). The most common grade 3–4 adverse events of special interest were cardiac disorders (41 [8%] of 542 patients in the abiraterone acetate group vs 20 [4%] of 540 patients in the placebo group), increased alanine aminotransferase (32 [6%] vs four [ vs 17 [3%]). Interpretation In this randomised phase 3 trial with a median follow-up of more than 4 years, treatment with abiraterone acetate prolonged overall survival compared with prednisone alone by a margin that was both clinically and statistically significant. These results further support the favourable safety profile of abiraterone acetate in patients with chemotherapy-naive metastatic castration-resistant prostate cancer. Funding Janssen Research & Development.

Abstract: Wireless communications is one of the most successful technologies in modern years, given that an exponential growth rate in wireless traffic has been sustained for over a century (known as Cooper’s law). This trend will certainly continue, driven by new innovative applications; for example, augmented reality and the Internet of Things. Massive MIMO has been identified as a key technology to handle orders of magnitude more data traffic. Despite the attention it is receiving from the communication community, we have personally witnessed that Massive MIMO is subject to several widespread misunderstandings, as epitomized by following (fictional) abstract: “The Massive MIMO technology uses a nearly infinite number of high-quality antennas at the base stations. By having at least an order of magnitude more antennas than active terminals, one can exploit asymptotic behaviors that some special kinds of wireless channels have. This technology looks great at first sight, but unfortunately the signal processing complexity is off the charts and the antenna arrays would be so huge that it can only be implemented in millimeter-wave bands.” These statements are, in fact, completely false. In this overview article, we identify 10 myths and explain why they are not true. We also ask a question that is critical for the practical adoption of the technology and which will require intense future research activities to answer properly. We provide references to key technical papers that support our claims, while a further list of related overview and technical papers can be found at the Massive MIMO Info Point: http://massivemimo. eu

Abstract: Mobile devices such as laptops, personal digital assistants, and mobile phones have become a learning tool with great potential in both classrooms and outdoor learning. Although there have been qualitative analyses of the use of mobile devices in education, systematic quantitative analyses of the effects of mobile-integrated education are lacking. This study performed a meta-analysis and research synthesis of the effects of integrated mobile devices in teaching and learning, in which 110 experimental and quasiexperimental journal articles published during the period 1993-2013 were coded and analyzed. Overall, there was a moderate mean effect size of 0.523 for the application of mobile devices to education. The effect sizes of moderator variables were analyzed and the advantages and disadvantages of mobile learning in different levels of moderator variables were synthesized based on content analyses of individual studies. The results of this study and their implications for both research and practice are discussed. This is a meta-analysis and research synthesis study for mobile-integrated education.110 published journal articles that were written over a 20-year period were coded and analyzed.The application of mobile devices to education has a moderate mean effect size.The effect sizes of moderator variables were analyzed.The benefits and drawbacks of mobile learning were synthesized.

Abstract: Description Update of the 2009 US Preventive Services Task Force (USPSTF) recommendation on screening for depression in adults. Methods The USPSTF reviewed the evidence on the benefits and harms of screening for depression in adult populations, including older adults and pregnant and postpartum women; the accuracy of depression screening instruments; and the benefits and harms of depression treatment in these populations. Population This recommendation applies to adults 18 years and older. Recommendation The USPSTF recommends screening for depression in the general adult population, including pregnant and postpartum women. Screening should be implemented with adequate systems in place to ensure accurate diagnosis, effective treatment, and appropriate follow-up. (B recommendation)

Abstract: Basic life support (BLS), advanced cardiovascular life support (ACLS), and post–cardiac arrest care are labels of convenience that each describe a set of skills and knowledge that are applied sequentially during the treatment of patients who have a cardiac arrest. There is overlap as each stage of care progresses to the next, but generally ACLS comprises the level of care between BLS and post–cardiac arrest care. ACLS training is recommended for advanced providers of both prehospital and in-hospital medical care. In the past, much of the data regarding resuscitation was gathered from out-of-hospital arrests, but in recent years, data have also been collected from in-hospital arrests, allowing for a comparison of cardiac arrest and resuscitation in these 2 settings. While there are many similarities, there are also some differences between in- and out-of-hospital cardiac arrest etiology, which may lead to changes in recommended resuscitation treatment or in sequencing of care. The consideration of steroid administration for in-hospital cardiac arrest (IHCA) versus out-of-hospital cardiac arrest (OHCA) is one such example discussed in this Part. The recommendations in this 2015 American Heart Association (AHA) Guidelines Update for Cardiopulmonary Resuscitation (CPR) and Emergency Cardiovascular Care (ECC) are based on an extensive evidence review process that was begun by the International Liaison Committee on Resuscitation (ILCOR) after the publication of the ILCOR 2010 International Consensus on Cardiopulmonary Resuscitation and Emergency Cardiovascular Care Science With Treatment Recommendations 1 and was completed in February 2015.2 In this in-depth evidence review process, the ILCOR task forces examined topics and then generated prioritized lists of questions for systematic review. Questions were first formulated in PICO (population, intervention, comparator, outcome) format,3 and then a search strategy and inclusion and exclusion criteria were defined and a search for relevant articles was performed. The evidence was evaluated by using …

Abstract: The 2019 novel coronavirus disease (COVID-19), with a starting point in China, has spread rapidly among people living in other countries, and is approaching approximately 34,986,502 cases worldwide according to the statistics of European Centre for Disease Prevention and Control. There are a limited number of COVID-19 test kits available in hospitals due to the increasing cases daily. Therefore, it is necessary to implement an automatic detection system as a quick alternative diagnosis option to prevent COVID-19 spreading among people. In this study, five pre-trained convolutional neural network based models (ResNet50, ResNet101, ResNet152, InceptionV3 and Inception-ResNetV2) have been proposed for the detection of coronavirus pneumonia infected patient using chest X-ray radiographs. We have implemented three different binary classifications with four classes (COVID-19, normal (healthy), viral pneumonia and bacterial pneumonia) by using 5-fold cross validation. Considering the performance results obtained, it has seen that the pre-trained ResNet50 model provides the highest classification performance (96.1% accuracy for Dataset-1, 99.5% accuracy for Dataset-2 and 99.7% accuracy for Dataset-3) among other four used models.

Abstract: Literature reviews establish the foundation of academic inquires. However, in the planning field, we lack rigorous systematic reviews. In this article, through a systematic search on the methodology of literature review, we categorize a typology of literature reviews, discuss steps in conducting a systematic literature review, and provide suggestions on how to enhance rigor in literature reviews in planning education and research.

Abstract: As deep nets are increasingly used in applications suited for mobile devices, a fundamental dilemma becomes apparent: the trend in deep learning is to grow models to absorb ever-increasing data set sizes; however mobile devices are designed with very little memory and cannot store such large models. We present a novel network architecture, HashedNets, that exploits inherent redundancy in neural networks to achieve drastic reductions in model sizes. HashedNets uses a low-cost hash function to randomly group connection weights into hash buckets, and all connections within the same hash bucket share a single parameter value. These parameters are tuned to adjust to the HashedNets weight sharing architecture with standard backprop during training. Our hashing procedure introduces no additional memory overhead, and we demonstrate on several benchmark data sets that HashedNets shrink the storage requirements of neural networks substantially while mostly preserving generalization performance.

Abstract: Importance A key factor in assessing the effectiveness and cost-effectiveness of antiretroviral therapy (ART) as a prevention strategy is the absolute risk of HIV transmission through condomless sex with suppressed HIV-1 RNA viral load for both anal and vaginal sex. Objective To evaluate the rate of within-couple HIV transmission (heterosexual and men who have sex with men [MSM]) during periods of sex without condoms and when the HIV-positive partner had HIV-1 RNA load less than 200 copies/mL. Design, Setting, and Participants The prospective, observational PARTNER (Partners of People on ART—A New Evaluation of the Risks) study was conducted at 75 clinical sites in 14 European countries and enrolled 1166 HIV serodifferent couples (HIV-positive partner taking suppressive ART) who reported condomless sex (September 2010 to May 2014). Eligibility criteria for inclusion of couple-years of follow-up were condomless sex and HIV-1 RNA load less than 200 copies/mL. Anonymized phylogenetic analysis compared couples’ HIV-1 polymerase and envelope sequences if an HIV-negative partner became infected to determine phylogenetically linked transmissions. Exposures Condomless sexual activity with an HIV-positive partner taking virally suppressive ART. Main Outcomes and Measures Risk of within-couple HIV transmission to the HIV-negative partner Results Among 1166 enrolled couples, 888 (mean age, 42 years [IQR, 35-48]; 548 heterosexual [61.7%] and 340 MSM [38.3%]) provided 1238 eligible couple-years of follow-up (median follow-up, 1.3 years [IQR, 0.8-2.0]). At baseline, couples reported condomless sex for a median of 2 years (IQR, 0.5-6.3). Condomless sex with other partners was reported by 108 HIV-negative MSM (33%) and 21 heterosexuals (4%). During follow-up, couples reported condomless sex a median of 37 times per year (IQR, 15-71), with MSM couples reporting approximately 22 000 condomless sex acts and heterosexuals approximately 36 000. Although 11 HIV-negative partners became HIV-positive (10 MSM; 1 heterosexual; 8 reported condomless sex with other partners), no phylogenetically linked transmissions occurred over eligible couple-years of follow-up, giving a rate of within-couple HIV transmission of zero, with an upper 95% confidence limit of 0.30/100 couple-years of follow-up. The upper 95% confidence limit for condomless anal sex was 0.71 per 100 couple-years of follow-up. Conclusions and Relevance Among serodifferent heterosexual and MSM couples in which the HIV-positive partner was using suppressive ART and who reported condomless sex, during median follow-up of 1.3 years per couple, there were no documented cases of within-couple HIV transmission (upper 95% confidence limit, 0.30/100 couple-years of follow-up). Additional longer-term follow-up is necessary to provide more precise estimates of risk.

Abstract: In December 2019 unexplained pneumonia cases were initially reported in Wuhan, China. The pathogen, a novel coronavirus named severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), was isolated from lower respiratory tract samples of infected patients and the resultant disease was termed as COVID-19 (Coronavirus Disease 2019)1 . By Feb 15, COVID-19 has rapidly spread throughout China and across the world, until a pandemic condition was announced by March 112 .

Abstract: The genomic revolution has fundamentally changed how we survey biodiversity on earth. High-throughput sequencing ("HTS") platforms now enable the rapid sequencing of DNA from diverse kinds of environmental samples (termed "environmental DNA" or "eDNA"). Coupling HTS with our ability to associate sequences from eDNA with a taxonomic name is called "eDNA metabarcoding" and offers a powerful molecular tool capable of noninvasively surveying species richness from many ecosystems. Here, we review the use of eDNA metabarcoding for surveying animal and plant richness, and the challenges in using eDNA approaches to estimate relative abundance. We highlight eDNA applications in freshwater, marine and terrestrial environments, and in this broad context, we distill what is known about the ability of different eDNA sample types to approximate richness in space and across time. We provide guiding questions for study design and discuss the eDNA metabarcoding workflow with a focus on primers and library preparation methods. We additionally discuss important criteria for consideration of bioinformatic filtering of data sets, with recommendations for increasing transparency. Finally, looking to the future, we discuss emerging applications of eDNA metabarcoding in ecology, conservation, invasion biology, biomonitoring, and how eDNA metabarcoding can empower citizen science and biodiversity education.

Abstract: Recently, researchers have made significant progress combining the advances in deep learning for learning feature representations with reinforcement learning. Some notable examples include training agents to play Atari games based on raw pixel data and to acquire advanced manipulation skills using raw sensory inputs. However, it has been difficult to quantify progress in the domain of continuous control due to the lack of a commonly adopted benchmark. In this work, we present a benchmark suite of continuous control tasks, including classic tasks like cart-pole swing-up, tasks with very high state and action dimensionality such as 3D humanoid locomotion, tasks with partial observations, and tasks with hierarchical structure. We report novel findings based on the systematic evaluation of a range of implemented reinforcement learning algorithms. Both the benchmark and reference implementations are released at https://github.com/rllab/rllab in order to facilitate experimental reproducibility and to encourage adoption by other researchers.

Abstract: The original COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) checklist was developed to assess the methodological quality of single studies on measurement properties of Patient-Reported Outcome Measures (PROMs). Now it is our aim to adapt the COSMIN checklist and its four-point rating system into a version exclusively for use in systematic reviews of PROMs, aiming to assess risk of bias of studies on measurement properties. For each standard (i.e., a design requirement or preferred statistical method), it was discussed within the COSMIN steering committee if and how it should be adapted. The adapted checklist was pilot-tested to strengthen content validity in a systematic review on the quality of PROMs for patients with hand osteoarthritis. Most important changes were the reordering of the measurement properties to be assessed in a systematic review of PROMs; the deletion of standards that concerned reporting issues and standards that not necessarily lead to biased results; the integration of standards on general requirements for studies on item response theory with standards for specific measurement properties; the recommendation to the review team to specify hypotheses for construct validity and responsiveness in advance, and subsequently the removal of the standards about formulating hypotheses; and the change in the labels of the four-point rating system. The COSMIN Risk of Bias checklist was developed exclusively for use in systematic reviews of PROMs to distinguish this application from other purposes of assessing the methodological quality of studies on measurement properties, such as guidance for designing or reporting a study on the measurement properties.

Abstract: The development of magnetic resonance imaging (MRI) techniques has defined modern neuroimaging. Since its inception, tens of thousands of studies using techniques such as functional MRI and diffusion weighted imaging have allowed for the non-invasive study of the brain. Despite the fact that MRI is routinely used to obtain data for neuroscience research, there has been no widely adopted standard for organizing and describing the data collected in an imaging experiment. This renders sharing and reusing data (within or between labs) difficult if not impossible and unnecessarily complicates the application of automatic pipelines and quality assurance protocols. To solve this problem, we have developed the Brain Imaging Data Structure (BIDS), a standard for organizing and describing MRI datasets. The BIDS standard uses file formats compatible with existing software, unifies the majority of practices already common in the field, and captures the metadata necessary for most common data processing operations.

Abstract: We propose a general framework called Network Dissection for quantifying the interpretability of latent representations of CNNs by evaluating the alignment between individual hidden units and a set of semantic concepts. Given any CNN model, the proposed method draws on a data set of concepts to score the semantics of hidden units at each intermediate convolutional layer. The units with semantics are labeled across a broad range of visual concepts including objects, parts, scenes, textures, materials, and colors. We use the proposed method to test the hypothesis that interpretability is an axis-independent property of the representation space, then we apply the method to compare the latent representations of various networks when trained to solve different classification problems. We further analyze the effect of training iterations, compare networks trained with different initializations, and measure the effect of dropout and batch normalization on the interpretability of deep visual representations. We demonstrate that the proposed method can shed light on characteristics of CNN models and training methods that go beyond measurements of their discriminative power.

Abstract: On April 13th, 2019, OpenAI Five became the first AI system to defeat the world champions at an esports game. The game of Dota 2 presents novel challenges for AI systems such as long time horizons, imperfect information, and complex, continuous state-action spaces, all challenges which will become increasingly central to more capable AI systems. OpenAI Five leveraged existing reinforcement learning techniques, scaled to learn from batches of approximately 2 million frames every 2 seconds. We developed a distributed training system and tools for continual training which allowed us to train OpenAI Five for 10 months. By defeating the Dota 2 world champion (Team OG), OpenAI Five demonstrates that self-play reinforcement learning can achieve superhuman performance on a difficult task.

Abstract: INTRODUCTION Genetic interactions occur when mutations in two or more genes combine to generate an unexpected phenotype. An extreme negative or synthetic lethal genetic interaction occurs when two mutations, neither lethal individually, combine to cause cell death. Conversely, positive genetic interactions occur when two mutations produce a phenotype that is less severe than expected. Genetic interactions identify functional relationships between genes and can be harnessed for biological discovery and therapeutic target identification. They may also explain a considerable component of the undiscovered genetics associated with human diseases. Here, we describe construction and analysis of a comprehensive genetic interaction network for a eukaryotic cell. RATIONALE Genome sequencing projects are providing an unprecedented view of genetic variation. However, our ability to interpret genetic information to predict inherited phenotypes remains limited, in large part due to the extensive buffering of genomes, making most individual eukaryotic genes dispensable for life. To explore the extent to which genetic interactions reveal cellular function and contribute to complex phenotypes, and to discover the general principles of genetic networks, we used automated yeast genetics to construct a global genetic interaction network. RESULTS We tested most of the ~6000 genes in the yeast Saccharomyces cerevisiae for all possible pairwise genetic interactions, identifying nearly 1 million interactions, including ~550,000 negative and ~350,000 positive interactions, spanning ~90% of all yeast genes. Essential genes were network hubs, displaying five times as many interactions as nonessential genes. The set of genetic interactions or the genetic interaction profile for a gene provides a quantitative measure of function, and a global network based on genetic interaction profile similarity revealed a hierarchy of modules reflecting the functional architecture of a cell. Negative interactions connected functionally related genes, mapped core bioprocesses, and identified pleiotropic genes, whereas positive interactions often mapped general regulatory connections associated with defects in cell cycle progression or cellular proteostasis. Importantly, the global network illustrates how coherent sets of negative or positive genetic interactions connect protein complex and pathways to map a functional wiring diagram of the cell. CONCLUSION A global genetic interaction network highlights the functional organization of a cell and provides a resource for predicting gene and pathway function. This network emphasizes the prevalence of genetic interactions and their potential to compound phenotypes associated with single mutations. Negative genetic interactions tend to connect functionally related genes and thus may be predicted using alternative functional information. Although less functionally informative, positive interactions may provide insights into general mechanisms of genetic suppression or resiliency. We anticipate that the ordered topology of the global genetic network, in which genetic interactions connect coherently within and between protein complexes and pathways, may be exploited to decipher genotype-to-phenotype relationships.

Abstract: BackgroundOn the basis of data from a phase 2 trial that compared the checkpoint inhibitor ipilimumab at doses of 0.3 mg, 3 mg, and 10 mg per kilogram of body weight in patients with advanced melanoma, this phase 3 trial evaluated ipilimumab at a dose of 10 mg per kilogram in patients who had undergone complete resection of stage III melanoma. MethodsAfter patients had undergone complete resection of stage III cutaneous melanoma, we randomly assigned them to receive ipilimumab at a dose of 10 mg per kilogram (475 patients) or placebo (476) every 3 weeks for four doses, then every 3 months for up to 3 years or until disease recurrence or an unacceptable level of toxic effects occurred. Recurrence-free survival was the primary end point. Secondary end points included overall survival, distant metastasis–free survival, and safety. ResultsAt a median follow-up of 5.3 years, the 5-year rate of recurrence-free survival was 40.8% in the ipilimumab group, as compared with 30.3% in the placebo group (hazard ratio ...

Abstract: When building a unified vision system or gradually adding new capabilities to a system, the usual assumption is that training data for all tasks is always available. However, as the number of tasks grows, storing and retraining on such data becomes infeasible. A new problem arises where we add new capabilities to a Convolutional Neural Network (CNN), but the training data for its existing capabilities are unavailable. We propose our Learning without Forgetting method, which uses only new task data to train the network while preserving the original capabilities. Our method performs favorably compared to commonly used feature extraction and fine-tuning adaption techniques and performs similarly to multitask learning that uses original task data we assume unavailable. A more surprising observation is that Learning without Forgetting may be able to replace fine-tuning with similar old and new task datasets for improved new task performance.

Abstract: The active sites over commercial copper/zinc oxide/aluminum oxide (Cu/ZnO/Al2O3) catalysts for carbon dioxide (CO2) hydrogenation to methanol, the Zn-Cu bimetallic sites or ZnO-Cu interfacial sites, have recently been the subject of intense debate. We report a direct comparison between the activity of ZnCu and ZnO/Cu model catalysts for methanol synthesis. By combining x-ray photoemission spectroscopy, density functional theory, and kinetic Monte Carlo simulations, we can identify and characterize the reactivity of each catalyst. Both experimental and theoretical results agree that ZnCu undergoes surface oxidation under the reaction conditions so that surface Zn transforms into ZnO and allows ZnCu to reach the activity of ZnO/Cu with the same Zn coverage. Our results highlight a synergy of Cu and ZnO at the interface that facilitates methanol synthesis via formate intermediates.

Abstract: We propose an end-to-end deep learning architecture that produces a 3D shape in triangular mesh from a single color image. Limited by the nature of deep neural network, previous methods usually represent a 3D shape in volume or point cloud, and it is non-trivial to convert them to the more ready-to-use mesh model. Unlike the existing methods, our network represents 3D mesh in a graph-based convolutional neural network and produces correct geometry by progressively deforming an ellipsoid, leveraging perceptual features extracted from the input image. We adopt a coarse-to-fine strategy to make the whole deformation procedure stable, and define various of mesh related losses to capture properties of different levels to guarantee visually appealing and physically accurate 3D geometry. Extensive experiments show that our method not only qualitatively produces mesh model with better details, but also achieves higher 3D shape estimation accuracy compared to the state-of-the-art.

Abstract: This article challenges Fixed Effects (FE) modelling as the ‘default’ for time-series-cross-sectional and panel data. Understanding different within- and between-effects is crucial when choosing modelling strategies. The downside of Random Effects (RE) modelling – correlated lower-level covariates and higher-level residuals – is omitted-variable bias, solvable with Mundlak’s (1978a) formulation. Consequently, RE can provide everything FE promises and more, as confirmed by Monte-Carlo simulations, which additionally show problems with Plumper and Troeger’s FE Vector Decomposition method when data are unbalanced. As well as incorporating time-invariant variables, RE models are readily extendable, with random coefficients, cross-level interactions, and complex variance functions. We argue not simply for technical solutions to endogeneity, but the substantive importance of context/heterogeneity, modelled using RE. The implications extend beyond political science, to all multilevel datasets. However, omitted variables could still bias estimated higher-level variable effects; as with any model, care is required in interpretation.

Abstract: A composite film of nickel sulfide/platinum/titanium foil (NiS/Pt/Ti) with low cost and high electrocatalytic activity was synthesized by the use of an in situ electropolymerization route and proposed as a counter electrode (CE) catalyst for flexible dye-sensitized solar cells (FDSSCs). The FDSSC with the NiS/Pt/Ti CE exhibited a comparable power conversion efficiency of 7.20% to the FDSSC with the platinum/titanium (Pt/Ti) CE showing 6.07%. The surface morphology of the NiS/Pt/Ti CE with one-dimensional (1D) structure is characterized by using the scanning electron microscopy (SEM). The NiS/Pt/Ti CE also displayed multiple electrochemical functions of excellent conductivity, great electrocatalytic ability for iodine/triiodine, and low charge transfer resistance of 2.61 ± 0.02 Ω cm2, which were characterized by using the cyclic voltammetry (CV), electrochemical impedance spectroscopy (EIS), and Tafel polarization plots. The photocurrent-photovoltage (J-V) character curves were further used to calculate the theoretical optical light performance parameters of the FDSSCs. It may be said that the NiS/Pt/Ti counter electrode is a promising catalytic material to replace the expensive platinum in FDSSCs.