University of Massachusetts Amherst

About: University of Massachusetts Amherst is a education organization based out in Amherst Center, Massachusetts, United States. It is known for research contribution in the topics: Population & Galaxy. The organization has 37274 authors who have published 83965 publications receiving 3834996 citations. The organization is also known as: UMass Amherst & Massachusetts State College.

Fully hardware-implemented memristor convolutional neural network

Abstract: Memristor-enabled neuromorphic computing systems provide a fast and energy-efficient approach to training neural networks1–4. However, convolutional neural networks (CNNs)—one of the most important models for image recognition5—have not yet been fully hardware-implemented using memristor crossbars, which are cross-point arrays with a memristor device at each intersection. Moreover, achieving software-comparable results is highly challenging owing to the poor yield, large variation and other non-ideal characteristics of devices6–9. Here we report the fabrication of high-yield, high-performance and uniform memristor crossbar arrays for the implementation of CNNs, which integrate eight 2,048-cell memristor arrays to improve parallel-computing efficiency. In addition, we propose an effective hybrid-training method to adapt to device imperfections and improve the overall system performance. We built a five-layer memristor-based CNN to perform MNIST10 image recognition, and achieved a high accuracy of more than 96 per cent. In addition to parallel convolutions using different kernels with shared inputs, replication of multiple identical kernels in memristor arrays was demonstrated for processing different inputs in parallel. The memristor-based CNN neuromorphic system has an energy efficiency more than two orders of magnitude greater than that of state-of-the-art graphics-processing units, and is shown to be scalable to larger networks, such as residual neural networks. Our results are expected to enable a viable memristor-based non-von Neumann hardware solution for deep neural networks and edge computing. A fully hardware-based memristor convolutional neural network using a hybrid training method achieves an energy efficiency more than two orders of magnitude greater than that of graphics-processing units.

The Calibration of Mid-Infrared Star Formation Rate Indicators

Abstract: With the goal of investigating the degree to which the MIR emission traces the SFR, we analyze Spitzer 8 and 24 μm data of star-forming regions in a sample of 33 nearby galaxies with available HST NICMOS images in the Paα (1.8756 μm) emission line. The galaxies are drawn from the SINGS sample and cover a range of morphologies and a factor ~10 in oxygen abundance. Published data on local low-metallicity starburst galaxies and LIRGs are also included in the analysis. Both the stellar continuum-subtracted 8 μm emission and the 24 μm emission correlate with the extinction-corrected Paα line emission, although neither relationship is linear. Simple models of stellar populations and dust extinction and emission are able to reproduce the observed nonlinear trend of the 24 μm emission versus number of ionizing photons, including the modest deficiency of 24 μm emission in the low-metallicity regions, which results from a combination of decreasing dust opacity and dust temperature at low luminosities. Conversely, the trend of the 8 μm emission as a function of the number of ionizing photons is not well reproduced by the same models. The 8 μm emission is contributed, in larger measure than the 24 μm emission, by dust heated by nonionizing stellar populations, in addition to the ionizing ones, in agreement with previous findings. Two SFR calibrations, one using the 24 μm emission and the other using a combination of the 24 μm and Hα luminosities (Kennicutt and coworkers), are presented. No calibration is presented for the 8 μm emission because of its significant dependence on both metallicity and environment. The calibrations presented here should be directly applicable to systems dominated by ongoing star formation.

Chapter 19 DNA Transformation

Abstract: Publisher Summary This chapter discusses DNA transformation. DNA transformation assays in a whole organism provide experimental links between molecular structure and phenotype. Experiments with transgenic Caenorhabditis elegans start in general with the injection of DNA into the adult gonad. Effects on phenotype or gene expression patterns can be analyzed either in F1 progeny derived from the injected animals or in derived transgenic lines. Germ-line transformation has been achieved by microinjection of DNA directly into oocyte nuclei or by microinjection of DNA into the cytoplasm of the hermaphrodite syncytial gonad. Three forms of heritable DNA transformation have been observed in C. elegans are: (1) extra chromosomal transformation; (2) non-homologous integration; and (3) homologous integration. Setting up microinjection in a laboratory already equipped for C. elegans genetics and molecular biology requires a modest investment in space and money. A separate easily scoreable marker gene to identify transformed animals can be extremely useful in a variety of injection experiments. The propensity for injected DNA molecules to recombine with each other generally allows one to coinject the selectable marker with a DNA segment to be tested for activity.

Rapid Cenozoic glaciation of Antarctica induced by declining atmospheric CO2.

Abstract: The sudden, widespread glaciation of Antarctica and the associated shift towards colder temperatures at the Eocene/Oligocene boundary (approximately 34 million years ago) (refs 1-4) is one of the most fundamental reorganizations of global climate known in the geologic record. The glaciation of Antarctica has hitherto been thought to result from the tectonic opening of Southern Ocean gateways, which enabled the formation of the Antarctic Circumpolar Current and the subsequent thermal isolation of the Antarctic continent. Here we simulate the glacial inception and early growth of the East Antarctic Ice Sheet using a general circulation model with coupled components for atmosphere, ocean, ice sheet and sediment, and which incorporates palaeogeography, greenhouse gas, changing orbital parameters, and varying ocean heat transport. In our model, declining Cenozoic CO2 first leads to the formation of small, highly dynamic ice caps on high Antarctic plateaux. At a later time, a CO2 threshold is crossed, initiating ice-sheet height/mass-balance feedbacks that cause the ice caps to expand rapidly with large orbital variations, eventually coalescing into a continental-scale East Antarctic Ice Sheet. According to our simulation the opening of Southern Ocean gateways plays a secondary role in this transition, relative to CO2 concentration.

Reversible Hydrogels from Self-Assembling Artificial Proteins

Abstract: Recombinant DNA methods were used to create artificial proteins that undergo reversible gelation in response to changes in pH or temperature. The proteins consist of terminal leucine zipper domains flanking a central, flexible, water-soluble polyelectrolyte segment. Formation of coiled-coil aggregates of the terminal domains in near-neutral aqueous solutions triggers formation of a three-dimensional polymer network, with the polyelectrolyte segment retaining solvent and preventing precipitation of the chain. Dissociation of the coiled-coil aggregates through elevation of pH or temperature causes dissolution of the gel and a return to the viscous behavior that is characteristic of polymer solutions. The mild conditions under which gel formation can be controlled (near-neutral pH and near-ambient temperature) suggest that these materials have potential in bioengineering applications requiring encapsulation or controlled release of molecular and cellular species.