UC San Diego

Deep learning (DL) is one of the fastest-growing topics in materials data science, with rapidly emerging applications spanning atomistic, image-based, spectral, and textual data modalities. DL allows analysis of unstructured data and automated identification of features. The recent development of large materials databases has fueled the application of DL methods in atomistic prediction in particular. In contrast, advances in image and spectral data have largely leveraged synthetic data enabled by high-quality forward models as well as by generative unsupervised DL methods. In this article, we present a high-level overview of deep learning methods followed by a detailed discussion of recent developments of deep learning in atomistic simulation, materials imaging, spectral analysis, and natural language processing. For each modality we discuss applications involving both theoretical and experimental data, typical modeling approaches with their strengths and limitations, and relevant publicly available software and datasets. We conclude the review with a discussion of recent cross-cutting work related to uncertainty quantification in this field and a brief perspective on limitations, challenges, and potential growth areas for DL methods in materials science.

This paper describes how the contemporary technology innovation ecology is hostile to community-driven design. These hostilities are important to understand if we want to intervene in the policy landscape of technology innovation to support viable alternatives to big tech consolidation and more democratic ways of developing and maintaining technology. We contribute a thick description of the hostile ecologies faced by transportation workers, community organizers, and allied technology researchers as they work toward building a cooperatively-owned taxi business with a digital dispatching technology. Our findings show that the hostile innovation ecology manifests as constrained access to resources, an inequitable regulatory framework, diminished agency in the software design process, and limits to the will of our community partners. We discuss the paths toward innovation for United Taxi Workers San Diego as compared with transportation network companies (e.g. Lyft, Uber) in terms of access to funding, regulation, labor, expertise, and market. We argue that a critical examination of institutions and policies in the innovation ecology is a necessary step toward charting fair, equitable, and community-strengthening pathways for technology innovation in the future.

How much does specific previous experience shape immediate perception? Top-down perceptual inference occurs in ambiguous situations. However, similarity-based accounts such as exemplar theory suggest that similar memories resonate with the percept, predicting that detailed previous experiences can shape perception even when bottom-up cues are unambiguous. The current study tests whether specific musical memories influence beat perception only under ambiguity, or more pervasively-that is, even when clear bottom-up beat cues are present. Listeners were exposed to 16 melodies, half in one meter, half in another. Later, each listener's perception of a specific melody's beat pattern was tested when that melody occurred in either its original meter or another meter. Ratings of metrical probes were influenced not only by fit with the current (test) meter, but also by fit with the meter previously experienced with that melody. Findings suggest that perception is routinely influenced by detailed top-down perceptual imagery.

Sleep duration, sleep efficiency, and sleep timing have been shown to have potential effects on metabolic functions relevant to circadian rhythms. It is not clear if the impact of sleep patterns on metabolic risk factors is through sociocultural and environmental factors or circadian misalignment. We investigated the associations of sleep patterns, chronotype, and social jet lag with metabolic syndrome among non-shift worker Hispanic/Latino adults. We used cross-sectional data from the Sueño Ancillary Study of The Hispanic Community Health Study/Study of Latinos (HCHS/SOL). Data from a subsample of 2189 participants aged 18-64 years were used in the analysis. Mean nightly sleep duration, mean sleep onset time, mean sleep offset time, mean sleep midpoint time, sleep efficiency, sleep variability (standard deviation (SD) of sleep duration, and SD of sleep midpoint), and time spent above light exposure threshold (1000 lux) in a day were assessed by wrist actigraphy (Acti-watch Spectrum). Chronotype was determined by the reduced Morningness-Eveningness Questionnaire. Medical conditions including dyslipidemia, hypertension, and diabetes mellitus were determined from a fasting blood specimen and physical exam at the baseline visit. To determine whether sleep patterns, light levels, chronotype, and social jetlag are associated with metabolic syndrome, multivariable logistic regression models were fitted, including variables with P < .15 in the univariate analysis. The results of the multivariable analysis demonstrated that in participants older than 40 years, intermediate chronotype (vs early) was significantly associated with a higher risk of metabolic syndrome (Odds ratio (95%CI): 1.33 (1.04,1.7)), while later chronotype (vs. early) in participants younger than 40 years was significantly associated with a lower risk of metabolic syndrome (Odds ratio (95%CI): 0.37 (0.14, 0.96)). Also, higher sleep efficiency was significantly associated with decreased odds of metabolic syndrome (Odds ratio (95%CI): 0.98 (0.96, 0.99)). Nightly sleep duration was not significantly different between two groups of participants with and without metabolic syndrome in multivariable analyses. There was no significant association between social jet lag and metabolic syndrome in multivariable analysis (p = .286). Moreover, there was no significant association between chronotype and social jet lag in multivariable analysis. The association between metabolic syndrome and chronotype is age-dependent. While early chronotype is associated with metabolic syndrome in younger individuals, it tended to be associated with lower odds for metabolic syndrome in older individuals.

Abstract Collapse of the edge flow shear as the line-averaged density approaches the Greenwald density limit has been observed as a precursor to the enhanced edge particle flux characteristic of proximity to the density limit regime. Here, we report the use of a biased electrode to sustain the edge shear layer in high density discharges, in which the shear layer would otherwise collapse. A stable increase in line-averaged density is observed along with a strong increase in edge density. These experiments were carried out on the J-TEXT tokamak. The Reynolds stress at the edge is enhanced, and the zonal flow sustained, while density perturbation levels, the flux of turbulence internal energy (i.e., turbulence spreading), and particle and heat flux all decrease significantly. Electron adiabaticity increases, and bias voltage modulation experiments show that an increase in the edge shear leads the increase in adiabaticity. These results suggest that external edge E × B flow shear drive may be of interest for sustaining edge plasma states at high density, and support the hypothesis that collapse of the edge shear layer triggers the onset of the strong transport and turbulence characteristic of the density limit regime.

Objectives

Methods

Results

Conclusions

Background

Methods

Results

Conclusions

Abstract The Arctic seasonal halocline impacts the exchange of heat, energy, and nutrients between the surface and the deeper ocean, and it is changing in response to Arctic sea ice melt over the past several decades. Here, we assess seasonal halocline formation in 1975 and 2006–12 by comparing daily, May–September, salinity profiles collected in the Canada Basin under sea ice. We evaluate differences between the two time periods using a one-dimensional (1D) bulk model to quantify differences in freshwater input and vertical mixing. The 1D metrics indicate that two separate factors contribute similarly to stronger stratification in 2006–12 relative to 1975: 1) larger surface freshwater input and 2) less vertical mixing of that freshwater. The larger freshwater input is mainly important in August–September, consistent with a longer melt season in recent years. The reduced vertical mixing is mainly important from June until mid-August, when similar levels of freshwater input in 1975 and 2006–12 are mixed over a different depth range, resulting in different stratification. These results imply that decadal changes to ice–ocean dynamics, in addition to freshwater input, significantly contribute to the stronger seasonal stratification in 2006–12 relative to 1975. These findings highlight the need for near-surface process studies to elucidate the impact of lateral processes and ice–ocean momentum exchange on vertical mixing. Moreover, the results may provide insight for improving the representation of decadal changes to Arctic upper-ocean stratification in climate models that do not capture decadal changes to vertical mixing.

Aim

Methods

Results

Conclusion