Search

• Analog techniques can accelerate machine learning by orders of magnitude. • Multiple strategies, by analog designers, bypass limitations of analog devices. • Examples are based on current work at IBM. Digital logic technology has been extraordinarily successful and has been fueled by incredible gains in integration density and performance achieved over the years following Moore's law. This has led to societal changes where more and more everyday functions are aided by smart devices following the path to artificial intelligence. However, in the field of deep machine learning, even this technology falls short, partly because device scaling gains are no longer easy to come by, but also due to intractable energy costs of computation. Deep learning, using labeled data, can be mapped onto artificial neural networks, arrays where the inputs and outputs are connected by programmable weights, and which can perform pattern recognition functions. The learning process consists of finding the optimum weights however this learning process is very slow for large problems. Exploiting the fact that weights do not need to be determined with high precision, as long as they can be updated precisely, the device community has recognized that analog computation approaches, using physical arrays of memristor (programmable resistor) type devices could offer significant speedup and power advantages compared to pure digital, or pure software approaches. On the other hand, the history of analog computation is not reassuring since the rule has been that more capable digital devices invariably supplant analog function. In this paper I will discuss the opportunities and limitations of using analog techniques to accelerate the learning process in resistive neural networks. [ABSTRACT FROM AUTHOR]

Abstract: Dissertation advising is an important form of mentoring. To investigate the impact of dissertation advising over time, advisor and committee member names were collected for 2400 dissertations completed over a 40-year period (1964–2004) in 32 North American information and library science schools. Several mentoring impact metrics are reported for a subset of the data, including the number of dissertations advised, the number of dissertation committees served on, the ratio of advising to committee membership, and the fractional “mpact” that weights advising and committee membership. The subset consists of data for six schools that produced at least three dozen dissertations and for which complete data is available. The data and resulting “mpact” metrics offer new ways to assess faculty impact and to investigate the nature and growth of a field. [Copyright &y& Elsevier]

Bromodomains (BDs) regulate gene expression by recognizing protein motifs containing acetyllysine. Although originally characterized as histone-binding proteins, it has since become clear that these domains interact with other acetylated proteins, perhaps most prominently transcription factors. The likely transient nature and low stoichiometry of such modifications, however, has made it challenging to fully define the interactome of any given BD. To begin to address this knowledge gap in an unbiased manner, we carried out mRNA display screens against a BD--the N-terminal BD of BRD3--using peptide libraries that contained either one or two acetyllysine residues. We discovered peptides with very strong consensus sequences and with affinities that are significantly higher than typical BD-peptide interactions. X-ray crystal structures also revealed modes of binding that have not been seen with natural ligands. Intriguingly, however, our sequences are not found in the human proteome, perhaps suggesting that strong binders to BDs might have been selected against during evolution. [ABSTRACT FROM AUTHOR]

Central Asia is dominated by an arid climate and desert-like conditions, leading to the potential for long-range transport of desert dust within and out of the region. Of particular interest is the Aral Sea, which has receded in size largely due to water diversion. As a result, newly exposed sediments are resuspended by wind and thus, may be a potential new source of particulate matter within the region. Here, strontium and lead stable isotope ratios are employed along with detailed elemental composition, to explore the contribution of long-range transport of Aral Sea sediments, as well as other potential sources of dust, within Central Asia. Ambient PM 10 samples were collected during dust and non-dust events from mid-2008 to mid-2009 at two sites in Kyrgyzstan located ∼1200 and 1500 km ESE of the Aral Sea. Aral Sea sediments and local Kyrgyzstan soils were resuspended and sized to PM 10 . The Aral Sea sediments have an average 87 Sr/ 86 Sr ratio of 0.70992. In contrast, the Sr isotope ratio in local soils exhibits an average ratio of 0.71579. Ambient PM 10 collected in Kyrgyzstan has an average 87 Sr/ 86 Sr ratio of 0.71177, falling between the values of these two potential sources and indicating a complex mixture of contributing sources. At both sites, airborne Sr isotope ratios measured during dust events were similar, suggesting that Aral Sea sediments only minimally affect air quality in Kyrgyzstan. Elemental analysis and Pb isotope ratios supported this finding. While the Pb isotopes and elemental data both indicate an anthropogenic source, long-range dust transport from other deserts inside and outside the region cannot be ruled out as sources of PM 10 in Central Asia. [ABSTRACT FROM AUTHOR]

Dust storms are significant contributors to ambient levels of particulate matter (PM) in many areas of the world. Central Asia, an area that is relatively understudied in this regard, is anticipated to be affected by dust storms due to its proximity to several major deserts that are in and generally surround Central Asia (e.g., the Aral Sea region, the Taklimakan desert in Western China). To investigate the relative importance of mineral dust (dust specifically composed of soil related minerals and oxides) in Central Asia, PM 10 and PM 2.5 , and by difference, coarse particles (particles with diameters between 2.5 and 10 μm) were measured at two sites, Bishkek and Lidar Station Teplokluchenka (Lidar), in the Kyrgyz Republic. Samples were collected every other day from July 2008 to July 2009. Daily samples were analyzed for mass and organic and elemental carbon. Samples were also composited on a bi-weekly basis and analyzed for elemental constituents and ionic components. In addition, samples collected on days with relatively high and low PM concentrations were analyzed before, and separately, from the biweekly composites to investigate the chemical differences between the episodic events. Data from the episodic samples were averaged into the composited averages. Using the elemental component data, several observational models were examined to estimate the contribution of mineral dust to ambient PM levels. A mass balance was also conducted. Results indicate that at both sites, mineral dust (as approximated by the “dust oxide” model) and organic matter (OM) were the dominant contributors to PM 10 and PM 2.5 . Mineral dust was a more significant contributor to the coarse PM (PM 10-2.5 ) during high event samples at both sites, although the relative contribution is greater at the Lidar site (average ± standard deviation = 42 ± 29%) as compared with the Bishkek site (26 ± 16%). Principal Components Analysis (PCA) was performed using data from both sites, and PCA indicated that mineral dust explained the majority of the variance in PM concentrations, and that the major apportioned factors of PM 10 and PM 2.5 were chemically similar between sites. [ABSTRACT FROM AUTHOR]

Abstract: Central Asia is a relatively understudied region of the world in terms of characterizing ambient particulate matter (PM) and quantifying source impacts of PM at receptor locations, although it is speculated to have an important role as a source region for long-range transport of PM to Eastern Asia, the Pacific Ocean, and the Western United States. PM is of significant interest not only because of its adverse effect on public health but also due to its more recently realized role in climate change. To investigate the sources and characteristics of PM in the region, a series of PM2.5 and PM10 samples were collected on an every-other-day basis at two sites (termed “Bishkek” and “Teploklyuchenka”) in the Central Asian nation of the Kyrgyz Republic (also known as Kyrgyzstan) for a full year from July 2008 to July 2009. These samples were analyzed using standard methods for mass, organic carbon (OC), elemental carbon (EC), water-soluble organic carbon (WSOC), water-insoluble organic carbon by difference (OC minus WSOC) and a variety of molecular marker chemical species to be used in a chemical mass balance (CMB) model to apportion the sources of OC. These analyses indicate that approximately 19 ± 6.4% of the PM2.5 mass at both sites throughout the year consists of OC. The carbonaceous component of PM2.5 is dominated by OC, with OC/Total Carbon (TC) ratios being around 0.8 in the winter to almost 0.95 in the summer months. The CMB analysis indicated that mobile sources, i.e., gasoline and diesel engine exhaust, biomass combustion, and biogenic secondary organic aerosol (SOA) formation from isoprene and α-pinene precursors in the summer months were the dominant sources of OC. A strong positive correlation was observed between non-biomass burning WSOC and the un-apportioned OC from the CMB analysis, indicating that some of this un-apportioned OC is WSOC and likely the result of SOA-forming atmospheric processes that were not estimated by the CMB analysis performed. In addition, a comparison of the predominant contributors to OC between the two sites indicates that biomass combustion is a stronger relative source of OC at the Teploklyuchenka site, particularly in the winter, while contributions of isoprene- and α-pinene-derived SOA to the measured OC was relatively similar between the sites. [Copyright &y& Elsevier]

Abstract: This study presents the development and field evaluation of a compact high-volume dichotomous sampler (HVDS) that collects coarse and fine particulate matter. The key features of this device is the utilization of a round nozzle virtual impactor with a 50% cutpoint at aerodynamic diameter, operating at a high inlet flow rate with a calculated Reynolds number of approximately . In its primary configuration as tested, the sampler size-fractionates into coarse and fine fractions with a minor flow ratio of 10% (minor/total flow rate), with major and minor flow rates of 900 and , respectively. Performance evaluation for concentration enrichment was conducted with a 4% minor flow ratio ( minor flow) as well. Tests demonstrated near ideal results at both 10% and 4% minor flow ratios, indicating enrichment to be independent of minor flow rates within the range evaluated. Reasonable agreement was found between the new sampler and collocated Partisol and MOUDI for ambient measurements. A tandem virtual impactor super-concentration system was also tested with the virtual impactor followed by a round nozzle virtual impactor and proved to be an efficient system to achieve higher concentration enrichment of ambient particles up to 180 times ambient levels. Investigation of the effect of different ambient parameters like RH and wind speed on coarse PM concentration corroborates results from earlier studies. The HVDS is an effective system to collect coarse and fine PM simultaneously, allowing for comprehensive standard chemical analyses over short sampling intervals. [Copyright &y& Elsevier]