skip to main content


Title: Bolstering geometric morphometrics sample sizes with damaged and pathologic specimens: Is near enough good enough?
Award ID(s):
1551766
NSF-PAR ID:
10301794
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Journal of Anatomy
Volume:
238
Issue:
6
ISSN:
0021-8782
Page Range / eLocation ID:
1444 to 1455
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. null (Ed.)
    ABSTRACT Salt-marsh foraminifera are sea-level proxies used to quantitatively reconstruct Holocene paleo-marsh elevations (PME) and subsequently relative sea level (RSL). The reliability of these reconstructions is partly dependent upon counting enough foraminifera to accurately characterize assemblages, while counting fewer tests allows more samples to be processed. We test the influence of count size on PME reconstructions by repeatedly subsampling foraminiferal assemblages preserved in a core of salt-marsh peat (from Newfoundland, Canada) with unusually large counts (up to 1595). Application of a single, weighted-averaging transfer function developed from a regional-scale modern training set to these ecologically-plausible simulated assemblages generated PME reconstructions at count sizes of 10–700. Reconstructed PMEs stabilize at counts sizes greater than ∼50 and counts exceeding ∼250 tests show little return for the additional time invested. The absence of some rare taxa in low counts is unlikely to markedly influence results from weighted-averaging transfer functions. Subsampling of modern foraminifera indicates that cross-validated transfer function performance shows only modest improvement when more than ∼40 foraminifera are counted. Studies seeking to understand multi-meter and millennial scale RSL trends should count more than ∼50 tests. The precision sought by studies aiming to resolve decimeter- and decadal-scale RSL variability is best achieved with counts greater than ∼75. In most studies seeking to reconstruct PME, effort is more productively allocated by counting relatively fewer foraminifera in more core samples than in counting large numbers of individuals. Target count sizes of 100–300 in existing studies are likely conservative and robust. Given the low diversity of salt-marsh foraminiferal assemblages, our results are likely applicable throughout and beyond northeastern North America. 
    more » « less
  2. Convolutional neural networks (CNNs) are used in numerous real-world applications such as vision-based autonomous driving and video content analysis. To run CNN inference on various target devices, hardware-aware neural architecture search (NAS) is crucial. A key requirement of efficient hardware-aware NAS is the fast evaluation of inference latencies in order to rank different architectures. While building a latency predictor for each target device has been commonly used in state of the art, this is a very time-consuming process, lacking scalability in the presence of extremely diverse devices. In this work, we address the scalability challenge by exploiting latency monotonicity --- the architecture latency rankings on different devices are often correlated. When strong latency monotonicity exists, we can re-use architectures searched for one proxy device on new target devices, without losing optimality. In the absence of strong latency monotonicity, we propose an efficient proxy adaptation technique to significantly boost the latency monotonicity. Finally, we validate our approach and conduct experiments with devices of different platforms on multiple mainstream search spaces, including MobileNet-V2, MobileNet-V3, NAS-Bench-201, ProxylessNAS and FBNet. Our results highlight that, by using just one proxy device, we can find almost the same Pareto-optimal architectures as the existing per-device NAS, while avoiding the prohibitive cost of building a latency predictor for each device. 
    more » « less
  3. null (Ed.)
  4. It remains unclear what sets the efficiency with which molecular gas transforms into stars. Here we present a new VLA map of the spiral galaxy M 51 in 33 GHz radio continuum, an extinction-free tracer of star formation, at 3″ scales (∼100 pc). We combined this map with interferometric PdBI/NOEMA observations of CO(1–0) and HCN(1–0) at matched resolution for three regions in M 51 (central molecular ring, northern and southern spiral arm segments). While our measurements roughly fall on the well-known correlation between total infrared and HCN luminosity, bridging the gap between Galactic and extragalactic observations, we find systematic offsets from that relation for different dynamical environments probed in M 51; for example, the southern arm segment is more quiescent due to low star formation efficiency (SFE) of the dense gas, despite its high dense gas fraction. Combining our results with measurements from the literature at 100 pc scales, we find that the SFE of the dense gas and the dense gas fraction anti-correlate and correlate, respectively, with the local stellar mass surface density. This is consistent with previous kpc-scale studies. In addition, we find a significant anti-correlation between the SFE and velocity dispersion of the dense gas. Finally, we confirm that a correlation also holds between star formation rate surface density and the dense gas fraction, but it is not stronger than the correlation with dense gas surface density. Our results are hard to reconcile with models relying on a universal gas density threshold for star formation and suggest that turbulence and galactic dynamics play a major role in setting how efficiently dense gas converts into stars. 
    more » « less