Search

Training neural networks with backpropagation (BP) requires a sequential passing of activations and gradients. This has been recognized as the lockings (i.e., the forward, backward, and update lockings) among modules (each module contains a stack of layers) inherited from the BP. In this brief, we propose a fully decoupled training scheme using delayed gradients (FDG) to break all these lockings. The FDG splits a neural network into multiple modules and trains them independently and asynchronously using different workers (e.g., GPUs). We also introduce a gradient shrinking process to reduce the stale gradient effect caused by the delayed gradients. Our theoretical proofs show that the FDG can converge to critical points under certain conditions. Experiments are conducted by training deep convolutional neural networks to perform classification tasks on several benchmark data sets. These experiments show comparable or better results of our approach compared with the state-of-the-art methods in terms of generalization and acceleration. We also show that the FDG is able to train various networks, including extremely deep ones (e.g., ResNet-1202), in a decoupled fashion. [ABSTRACT FROM AUTHOR]

A recent report discusses research findings on biology, specifically focusing on hereditary spastic paraplegias (HSPs), a group of inherited neurological disorders. The primary manifestation of HSPs is impaired gait and leg spasticity. The study aimed to identify genetic modifiers of endoplasmic reticulum (ER) organization, as mutations in genes encoding ER-shaping proteins are a common cause of HSP. The researchers developed a quantitative image-based screen to examine ER distribution in cells and identified significant changes in ER organization caused by the knockdown of HSP-causing genes ATL1 and RTN2. This study provides valuable insights into the cellular mechanisms underlying HSPs. [Extracted from the article]

In the 144 years since 'On the Origin of Species,' evolutionary theory has come a long way. Modern biology consistently confirms biologist, Charles Darwin's amazing ideas, even though he knew nothing of genetics or molecular biology. This article examines five of the biggest questions related to evolution. They are about origin of life, role of mutations in evolution, ways how new species are formed, whether evolution is predictable and the role of God in evolution.

The article discusses the role of mass extinctions in affecting the course of life of species. The author believes that there is not enough data to extrapolate rules why survivors of mass extinction do not last in the long term. He cites various species such as the multituberculates and the brachiopods which survived the asteroid impact and the Permian-Triassic extinction, respectively, but still died out or suffered decline. Despite insights from fossils, scientists do not have certain measures to predict future survivors of mass extinctions.

Scientists at the University of Guelph in Canada are using artificial intelligence (AI) to help identify and study bugs in order to better understand biodiversity. The project, called BugShot, uses deep learning, high-definition cameras, and computer vision to classify bugs from bulk samples, potentially reducing the identification process from a year to just a few hours. While AI has its limitations and requires human assistance, BugShot combines the AI expertise of the Vector Institute with the University of Guelph's extensive biodiversity data. The project aims to preserve biodiversity and address the potential threat of food web collapse caused by declining insect populations. [Extracted from the article]

Keywords for this news article include: AstraZeneca, Gothenburg, Sweden, Europe, Pharmaceutical Companies, Biology, Genetics, AstraZeneca, Life Sciences, Drugs and Therapies, Health and Medicine. Keywords: AstraZeneca; Biology; Business; Drugs and Therapies; Genetics; Health and Medicine; Life Sciences; Pharmaceutical Companies EN AstraZeneca Biology Business Drugs and Therapies Genetics Health and Medicine Life Sciences Pharmaceutical Companies 264 264 1 10/03/23 20231006 NES 231006 2023 OCT 6 (NewsRx) -- By a News Reporter-Staff News Editor at Drug Week -- Researchers detail new data in biology. [Extracted from the article]

SUMMARY: The dissection of apicomplexan biology has been highly influenced by the genetic tools available for manipulation of parasite DNA. Here, we describe different techniques available for the generation of conditional mutants. Comparison of the advantages and disadvantages of the three most commonly used regulation systems: the tetracycline inducible system, the regulation of protein stability and site-specific recombination are discussed. Using some previously described examples we explore some of the pitfalls involved in gene-function analysis using these systems that can lead to wrong or over-interpretation of phenotypes. We will also mention different options to standardize the application of these techniques for the characterization of gene function in high-throughput. [ABSTRACT FROM AUTHOR]

The article investigates on the Earth's past mass extinction events. Described by the author as much messier, the tale of mass extinction has revealed some rules that will tell about the future of life on the planet. He stresses that one species' survival does not guarantee the rise of another species. Though mass extinctions are often damaging, scientists revealed that they do not change the layout of life on the planet. Paleontologists cite the role of geographic range in extinction survival at the genus level and survivorship of larger groups.

Focuses on the significance of chemistry, biology and informatics tools in drug discovery. Enhancement of the structure-activity relationships capabilities; Predictions that provide directions to chemists; Major integration among traditional medical chemists; Demonstration of the possibility of differentiation between different groups of toxic compounds. INSET: From a Last Resort to Mainstream.

Does high familiarity with a face enable particularly efficient visual processing? In three experiments, we presented briefly and successively two pairs of faces (either famous or recently learned), masking each presentation. Between the first and the second presentations, one face changed, and the task was to locate this change. Performance was significantly better when the change involved a famous face. This superfamiliarity effect was found only for changes occurring in the left visual field and was abolished by inverting the faces. Extended prior study of the recently learned faces did not improve performance with these stimuli. The results suggest that superfamiliarity promotes highly efficient visual processing and may especially activate a configural mode of analysis. [ABSTRACT FROM AUTHOR]

The article discusses the status of genomics in India. Topics discussed include genomics research, genomics application to biology, and public-private partnerships. It features the insights of National Institute of Biomedical Genomics (NIBMG) director Partha Majumder, Xcode Life Sciences physician director Pankaj Mankad, and SciGenom chief executive officer (CEO) Sam Santhosh.

Discusses several issues related to biology as of June 2002. Concern on wildlife devastation in Zimbabwe; Significance of nutshells to biofuel crisis; Information on the report 'Review of Supply of Scientists and Engineers,' undertaken by Sir Gareth Roberts.