Hasil untuk "Architecture"

Dongwan Kim, Joo-Yeon Lee, Jeong-Sun Yang et al.

SARS-CoV-2 is a betacoronavirus that is responsible for the COVID-19 pandemic. The genome of SARS-CoV-2 was reported recently, but its transcriptomic architecture is unknown. Utilizing two complementary sequencing techniques, we here present a high-resolution map of the SARS-CoV-2 transcriptome and epitranscriptome. DNA nanoball sequencing shows that the transcriptome is highly complex owing to numerous recombination events, both canonical and noncanonical. In addition to the genomic RNA and subgenomic RNAs common in all coronaviruses, SARS-CoV-2 produces a large number of transcripts encoding unknown ORFs with fusion, deletion, and/or frameshift. Using nanopore direct RNA sequencing, we further find at least 41 RNA modification sites on viral transcripts, with the most frequent motif being AAGAA. Modified RNAs have shorter poly(A) tails than unmodified RNAs, suggesting a link between the internal modification and the 3′ tail. Functional investigation of the unknown ORFs and RNA modifications discovered in this study will open new directions to our understanding of the life cycle and pathogenicity of SARS-CoV-2. Highlights We provide a high-resolution map of SARS-CoV-2 transcriptome and epitranscriptome using nanopore direct RNA sequencing and DNA nanoball sequencing. The transcriptome is highly complex owing to numerous recombination events, both canonical and noncanonical. In addition to the genomic and subgenomic RNAs common in all coronaviruses, SARS-CoV-2 produces transcripts encoding unknown ORFs. We discover at least 41 potential RNA modification sites with an AAGAA motif.

Nabil Ibtehaz, Mohammad Sohel Rahman

In recent years Deep Learning has brought about a breakthrough in Medical Image Segmentation. In this regard, U-Net has been the most popular architecture in the medical imaging community. Despite outstanding overall performance in segmenting multimodal medical images, through extensive experimentations on some challenging datasets, we demonstrate that the classical U-Net architecture seems to be lacking in certain aspects. Therefore, we propose some modifications to improve upon the already state-of-the-art U-Net model. Following these modifications, we develop a novel architecture, MultiResUNet, as the potential successor to the U-Net architecture. We have tested and compared MultiResUNet with the classical U-Net on a vast repertoire of multimodal medical images. Although only slight improvements in the cases of ideal images are noticed, remarkable gains in performance have been attained for the challenging ones. We have evaluated our model on five different datasets, each with their own unique challenges, and have obtained a relative improvement in performance of 10.15%, 5.07%, 2.63%, 1.41%, and 0.62% respectively. We have also discussed and highlighted some qualitatively superior aspects of MultiResUNet over classical U-Net that are not really reflected in the quantitative measures.

J. Sweller, J. V. van Merriënboer, F. Paas

Cognitive load theory was introduced in the 1980s as an instructional design theory based on several uncontroversial aspects of human cognitive architecture. Our knowledge of many of the characteristics of working memory, long-term memory and the relations between them had been well-established for many decades prior to the introduction of the theory. Curiously, this knowledge had had a limited impact on the field of instructional design with most instructional design recommendations proceeding as though working memory and long-term memory did not exist. In contrast, cognitive load theory emphasised that all novel information first is processed by a capacity and duration limited working memory and then stored in an unlimited long-term memory for later use. Once information is stored in long-term memory, the capacity and duration limits of working memory disappear transforming our ability to function. By the late 1990s, sufficient data had been collected using the theory to warrant an extended analysis resulting in the publication of Sweller et al. (Educational Psychology Review, 10, 251–296, 1998). Extensive further theoretical and empirical work have been carried out since that time and this paper is an attempt to summarise the last 20 years of cognitive load theory and to sketch directions for future research.

Han Cai, Ligeng Zhu, Song Han

Neural architecture search (NAS) has a great impact by automatically designing effective neural network architectures. However, the prohibitive computational demand of conventional NAS algorithms (e.g. $10^4$ GPU hours) makes it difficult to \emph{directly} search the architectures on large-scale tasks (e.g. ImageNet). Differentiable NAS can reduce the cost of GPU hours via a continuous representation of network architecture but suffers from the high GPU memory consumption issue (grow linearly w.r.t. candidate set size). As a result, they need to utilize~\emph{proxy} tasks, such as training on a smaller dataset, or learning with only a few blocks, or training just for a few epochs. These architectures optimized on proxy tasks are not guaranteed to be optimal on the target task. In this paper, we present \emph{ProxylessNAS} that can \emph{directly} learn the architectures for large-scale target tasks and target hardware platforms. We address the high memory consumption issue of differentiable NAS and reduce the computational cost (GPU hours and GPU memory) to the same level of regular training while still allowing a large candidate set. Experiments on CIFAR-10 and ImageNet demonstrate the effectiveness of directness and specialization. On CIFAR-10, our model achieves 2.08\% test error with only 5.7M parameters, better than the previous state-of-the-art architecture AmoebaNet-B, while using 6$\times$ fewer parameters. On ImageNet, our model achieves 3.1\% better top-1 accuracy than MobileNetV2, while being 1.2$\times$ faster with measured GPU latency. We also apply ProxylessNAS to specialize neural architectures for hardware with direct hardware metrics (e.g. latency) and provide insights for efficient CNN architecture design.

Muhan Zhang, Zhicheng Cui, Marion Neumann et al.

Neural networks are typically designed to deal with data in tensor forms. In this paper, we propose a novel neural network architecture accepting graphs of arbitrary structure. Given a dataset containing graphs in the form of (G,y) where G is a graph and y is its class, we aim to develop neural networks that read the graphs directly and learn a classification function. There are two main challenges: 1) how to extract useful features characterizing the rich information encoded in a graph for classification purpose, and 2) how to sequentially read a graph in a meaningful and consistent order. To address the first challenge, we design a localized graph convolution model and show its connection with two graph kernels. To address the second challenge, we design a novel SortPooling layer which sorts graph vertices in a consistent order so that traditional neural networks can be trained on the graphs. Experiments on benchmark graph classification datasets demonstrate that the proposed architecture achieves highly competitive performance with state-of-the-art graph kernels and other graph neural network methods. Moreover, the architecture allows end-to-end gradient-based training with original graphs, without the need to first transform graphs into vectors.

Amit Zeisel, Hannah Hochgerner, P. Lönnerberg et al.

The mammalian nervous system executes complex behaviors controlled by specialised, precisely positioned and interacting cell types. Here, we used RNA sequencing of half a million single cells to create a detailed census of cell types in the mouse nervous system. We mapped cell types spatially and derived a hierarchical, data-driven taxonomy. Neurons were the most diverse, and were grouped by developmental anatomical units, and by the expression of neurotransmitters and neuropeptides. Neuronal diversity was driven by genes encoding cell identity, synaptic connectivity, neurotransmission and membrane conductance. We discovered several distinct, regionally restricted, astrocytes types, which obeyed developmental boundaries and correlated with the spatial distribution of key glutamate and glycine neurotransmitters. In contrast, oligodendrocytes showed a loss of regional identity, followed by a secondary diversification. The resource presented here lays a solid foundation for understanding the molecular architecture of the mammalian nervous system, and enables genetic manipulation of specific cell types.

Chenxi Liu, Barret Zoph, Jonathon Shlens et al.

We propose a new method for learning the structure of convolutional neural networks (CNNs) that is more efficient than recent state-of-the-art methods based on reinforcement learning and evolutionary algorithms. Our approach uses a sequential model-based optimization (SMBO) strategy, in which we search for structures in order of increasing complexity, while simultaneously learning a surrogate model to guide the search through structure space. Direct comparison under the same search space shows that our method is up to 5 times more efficient than the RL method of Zoph et al. (2018) in terms of number of models evaluated, and 8 times faster in terms of total compute. The structures we discover in this way achieve state of the art classification accuracies on CIFAR-10 and ImageNet.

Ping Chi, Shuangchen Li, Conglei Xu et al.

C. Curtis, Sohrab P. Shah, S. Chin et al.

The elucidation of breast cancer subgroups and their molecular drivers requires integrated views of the genome and transcriptome from representative numbers of patients. We present an integrated analysis of copy number and gene expression in a discovery and validation set of 997 and 995 primary breast tumours, respectively, with long-term clinical follow-up. Inherited variants (copy number variants and single nucleotide polymorphisms) and acquired somatic copy number aberrations (CNAs) were associated with expression in ∼40% of genes, with the landscape dominated by cis- and trans-acting CNAs. By delineating expression outlier genes driven in cis by CNAs, we identified putative cancer genes, including deletions in PPP2R2A, MTAP and MAP2K4. Unsupervised analysis of paired DNA–RNA profiles revealed novel subgroups with distinct clinical outcomes, which reproduced in the validation cohort. These include a high-risk, oestrogen-receptor-positive 11q13/14 cis-acting subgroup and a favourable prognosis subgroup devoid of CNAs. Trans-acting aberration hotspots were found to modulate subgroup-specific gene networks, including a TCR deletion-mediated adaptive immune response in the ‘CNA-devoid’ subgroup and a basal-specific chromosome 5 deletion-associated mitotic network. Our results provide a novel molecular stratification of the breast cancer population, derived from the impact of somatic CNAs on the transcriptome.

Stephen M. Smith, Peter T. Fox, K. Miller et al.

Mohammad Al-Fares, Alexander Loukissas, Amin Vahdat

R. Collobert, J. Weston

G. Foschini

J. Sweller, J. V. van Merrienboer, F. Paas

L. Bass, P. Clements, R. Kazman

S. Blake, David L. Black, M. Carlson et al.

Jason L. Hill, R. Szewczyk, A. Woo et al.

H. Simon

P. Resnick

E. Rosen, A. Viswanathan, R. Callon