Search

A large percentage of medical information is in unstructured text format in electronic medical record systems. Manual extraction of information from clinical notes is extremely time consuming. Natural language processing has been widely used in recent years for automatic information extraction from medical texts. However, algorithms trained on data from a single healthcare provider are not generalizable and error-prone due to the heterogeneity and uniqueness of medical documents. We develop a two-stage federated natural language processing method that enables utilization of clinical notes from different hospitals or clinics without moving the data, and demonstrate its performance using obesity and comorbities phenotyping as medical task. This approach not only improves the quality of a specific clinical task but also facilitates knowledge progression in the whole healthcare system, which is an essential part of learning health system. To the best of our knowledge, this is the first application of federated machine learning in clinical NLP., Comment: 9 pages; Proceedings of the 18th BioNLP Workshop and Shared Task

Unsupervised domain adaptation (UDA) is the task of modifying a statistical model trained on labeled data from a source domain to achieve better performance on data from a target domain, with access to only unlabeled data in the target domain. Existing state-of-the-art UDA approaches use neural networks to learn representations that can predict the values of subset of important features called "pivot features." In this work, we show that it is possible to improve on these methods by jointly training the representation learner with the task learner, and examine the importance of existing pivot selection methods., Comment: To be presented at NAACL 2019

Electronic health record (EHR) data is collected by individual institutions and often stored across locations in silos. Getting access to these data is difficult and slow due to security, privacy, regulatory, and operational issues. We show, using ICU data from 58 different hospitals, that machine learning models to predict patient mortality can be trained efficiently without moving health data out of their silos using a distributed machine learning strategy. We propose a new method, called Federated-Autonomous Deep Learning (FADL) that trains part of the model using all data sources in a distributed manner and other parts using data from specific data sources. We observed that FADL outperforms traditional federated learning strategy and conclude that balance between global and local training is an important factor to consider when design distributed machine learning methods , especially in healthcare., Comment: Machine Learning for Health (ML4H) Workshop at NeurIPS 2018 arXiv:cs/0101200

The HIgh-Resolution Mid-infrarEd Spectrometer (HIRMES) is the 3rd Generation Instrument for the Stratospheric Observatory For Infrared Astronomy (SOFIA), currently in development at the NASA Goddard Space Flight Center (GSFC), and due for commissioning in 2019. By combining direct-detection Transition Edge Sensor (TES) bolometer arrays, grating-dispersive spectroscopy, and a host of Fabry-Perot tunable filters, HIRMES will provide the ability for High Resolution (R~100,000), Mid-Resolution (R~10,000), and Low-Resolution (R~600) slit-spectroscopy, and 2D Spectral Imaging (R~2000 at selected wavelengths) over the 25 - 122 {\mu}m mid-far infrared waveband. The driving science application is the evolution of proto-planetary systems via measurements of water-vapor, water-ice, deuterated hydrogen (HD), and neutral oxygen lines. However, HIRMES has been designed to be as flexible as possible to cover a wide range of science cases that fall within its phase-space, all whilst reaching sensitivities and observing powers not yet seen thus far on SOFIA, providing unique observing capabilities which will remain unmatched for decades., Comment: Accepted to the Journal of Astronomical Instrumentation (SOFIA Special Edition) on 11th November 2018

This paper represents the conclusion of a project that had two main goals: (1) to investigate to what extent planetary nebulae (PNe) are chemically homogeneous; and (2) to provide physical constraints on the central star properties of each PN. We accomplished the first goal by using HST/STIS spectra to measure the abundances of seven elements in numerous spatial regions within each of six PN (IC 2165, IC 3568, NGC 2440, NGC 5315, NGC 5882, and NGC 7662). The second goal was achieved by computing a photoionization model of each nebula, using our observed emission line strengths as constraints. The major finding of our study is that the nebular abundances of He, C, N, O, Ne, S, and Ar are consistent with a chemically homogeneous picture for each PN. Additionally, we found through experimenting with three different density profiles (constant, Gaussian, and Gaussian with a power-law) that the determination of the central star's temperature and luminosity is only slightly sensitive to the profile choice. Lastly, post-AGB evolutionary model predictions of temperature and luminosity available in the literature were plotted along with the values inferred from the photoionization model analysis to yield initial and final mass estimates of each central star., Comment: 33 pages, 10 tables (5 span 18 pages), 7 figures, accepted for publication in MNRAS

There have been several recent attempts to improve the accuracy of grammar induction systems by bounding the recursive complexity of the induction model (Ponvert et al., 2011; Noji and Johnson, 2016; Shain et al., 2016; Jin et al., 2018). Modern depth-bounded grammar inducers have been shown to be more accurate than early unbounded PCFG inducers, but this technique has never been compared against unbounded induction within the same system, in part because most previous depth-bounding models are built around sequence models, the complexity of which grows exponentially with the maximum allowed depth. The present work instead applies depth bounds within a chart-based Bayesian PCFG inducer (Johnson et al., 2007b), where bounding can be switched on and off, and then samples trees with and without bounding. Results show that depth-bounding is indeed significantly effective in limiting the search space of the inducer and thereby increasing the accuracy of the resulting parsing model. Moreover, parsing results on English, Chinese and German show that this bounded model with a new inference technique is able to produce parse trees more accurately than or competitively with state-of-the-art constituency-based grammar induction models., Comment: EMNLP 2018

We analyze new HST/COS spectra for two quasar absorption outflows seen in the quasi-stellar object LBQS 1206+1052. These data cover, for the first time, absorption troughs from $S_{IV}$, $Si_{II}$, and $P_V$. From the ratio of the $S_{IV}$* to $S_{IV}$ column densities, we measure the electron number density of the higher-velocity ($-1400$ km s$^{-1}$, v1400) outflow to be log($n_e$) = $4.23^{+0.09}_{-0.09}$ cm$^{-3}$ and constrain the lower-velocity ($-730$ km s$^{-1}$, v700) outflow to log($n_e$) $>$ $5.3$ cm$^{-3}$. The $n_e$ associated with the higher-velocity outflow is an order of magnitude larger than reported in prior work. We find that the previous measurement was unreliable since it was based on density-sensitive absorption troughs that were likely saturated. Using photoionization models, we determine the best $\chi^2$-minimization fit for the ionization parameter and hydrogen column density of the higher-velocity outflow: log($U_H$) = $-1.73^{+0.21}_{-0.12}$ and log($N_H$) = $21.03^{+0.25}_{-0.15}$ cm$^{-2}$, respectively. We calculate from $U_H$ and $n_e$ a distance of $500^{+100}_{-110}$ pc from the central source to the outflow. Using an SED attenuated by the v700 outflow yields a two-phase photoionization solution for the v1400 outflow, separated by a $\Delta U \approxeq 0.7$. Otherwise, the resultant distance, mass flux, and kinetic luminosity are similar to the unattenuated case. However, the attenuated analysis has significant uncertainties due to a lack of constraints on the v700 outflow in 2017., Comment: 11 pages, 6 figures, 4 tables, accepted for publication in ApJ

The Dark Energy Spectroscopic Instrument (DESI) is under construction to measure the expansion history of the Universe using the Baryon Acoustic Oscillation technique. The spectra of 35 million galaxies and quasars over 14000 square degrees will be measured during the life of the experiment. A new prime focus corrector for the KPNO Mayall telescope will deliver light to 5000 fiber optic positioners. The fibers in turn feed ten broad-band spectrographs. We describe the DESI corrector optics, a series of six fused silica and borosilicate lenses. The lens diameters range from 0.8 to 1.1 meters, and their weights 84 to 237 kg. Most lens surfaces are spherical, and two are challenging 10th-order polynomial aspheres. The lenses have been successfully polished and treated with an antireflection coating at multiple subcontractors, and are now being integrated into the DESI corrector barrel assembly at University College London. We describe the final performance of the lenses in terms of their various parameters, including surface figure, homogeneity, and others, and compare their final performance against the demanding DESI corrector requirements. Also we describe the reoptimization of the lens spacing in their corrector barrel after their final measurements are known. Finally we assess the performance of the corrector as a whole, compared to early budgeted estimates.

Mining electronic health records for patients who satisfy a set of predefined criteria is known in medical informatics as phenotyping. Phenotyping has numerous applications such as outcome prediction, clinical trial recruitment, and retrospective studies. Supervised machine learning for phenotyping typically relies on sparse patient representations such as bag-of-words. We consider an alternative that involves learning patient representations. We develop a neural network model for learning patient representations and show that the learned representations are general enough to obtain state-of-the-art performance on a standard comorbidity detection task., Comment: Accepted to *SEM 2018

We conducted a survey of quasar outflows using the VLT/X-shooter spectrograph. When choosing the 14 BAL and mini-BALs comprising this sample, the data did not cover the S IV and S IV* troughs, whose ratio can be used to determine the distance of the outflows from the central source (R). Therefore, this "Blind Survey" is unbiased towards a particular distance scale. Out of the eight outflows where R can be measured, six have R > 100 pc (spanning the range 100-4500 pc), one has R > 10 pc, and only one (at R < 60 pc) is compatible with a much smaller R scale. At least two of the outflows have a kinetic luminosity greater than 0.5% of their Eddington luminosity, implying that they are able to provide significant AGN feedback. The outflows span a range of 0 to -10000 km s$^{-1}$ in velocity; total column density between 10$^{20}$ - 10$^{22.5}$ cm$^{-2}$ ; ionization parameter ($U_H$ ) in the range 0.01 - 1; and electron number density between 10$^3$ - 10$^{5.5}$ cm$^{-3}$ , with one upper and one lower limit. The results of this survey can be extrapolated to the majority of BAL outflows, implying that most of these outflows are situated far away from the AGN accretion disk; and that a significant portion of them can contribute to AGN feedback processes., Comment: 8 pages, 4 figures, 3 tables, submitted to ApJ

We determine the physical conditions and location of the outflow material seen in the mini-BAL quasar SDSS J1111+1437 (z = 2.138). These results are based on the analysis of a high S/N, medium-resolution VLT/X- shooter spectrum. The main outflow component spans the velocity range -1500 to -3000 km s$^{-1}$ and has detected absorption troughs from both high-ionization species: C IV , N V , O VI , Si IV , P V , and S IV ; and low-ionization species: H I , C II , Mg II , Al II , Al III , Si II , and Si III . Measurements of these troughs allow us to derive an accurate photoionization solution for this absorption component: a hydrogen column density, log ( N H ) = 21.47 $^{+0.21}_{-0.27}$ cm$^{-2}$ and ionization parameter, log (U_H) = -1.23$^{+0.20}_{-0.25}$. Troughs produced from the ground and excited states of S IV combined with the derived U_H value allow us to determine an electron number density of log (n_e) = 3.62$^{+0.09}_{-0.11}$ cm$^{-3}$ and to obtain the distance of the ionized gas from the central source: R = 880$^{+220}_{-260}$ pc., Comment: 9 pages, 6 figures, 2 tables. Published, 2018, ApJ, 858, 39

There has been recent interest in applying cognitively or empirically motivated bounds on recursion depth to limit the search space of grammar induction models (Ponvert et al., 2011; Noji and Johnson, 2016; Shain et al., 2016). This work extends this depth-bounding approach to probabilistic context-free grammar induction (DB-PCFG), which has a smaller parameter space than hierarchical sequence models, and therefore more fully exploits the space reductions of depth-bounding. Results for this model on grammar acquisition from transcribed child-directed speech and newswire text exceed or are competitive with those of other models when evaluated on parse accuracy. Moreover, gram- mars acquired from this model demonstrate a consistent use of category labels, something which has not been demonstrated by other acquisition models., Comment: Accepted by Transactions of the Association for Computational Linguistics