skip to main content


Search for: All records

Creators/Authors contains: "Treister, E"

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. Deep convolutional neural networks have revolutionized many machine learning and computer vision tasks, however, some remaining key challenges limit their wider use. These challenges include improving the network's robustness to perturbations of the input image and the limited ``field of view'' of convolution operators. We introduce the IMEXnet that addresses these challenges by adapting semi-implicit methods for partial differential equations. Compared to similar explicit networks, such as residual networks, our network is more stable, which has recently shown to reduce the sensitivity to small changes in the input features and improve generalization. The addition of an implicit step connects all pixels in each channel of the image and therefore addresses the field of view problem while still being comparable to standard convolutions in terms of the number of parameters and computational complexity. We also present a new dataset for semantic segmentation and demonstrate the effectiveness of our architecture using the NYU Depth dataset. 
    more » « less
  2. Convolutional Neural Networks (CNNs) filter the input data using spatial convolution operators with compact stencils. Commonly, the convolution operators couple features from all channels, which leads to immense computational cost in the training of and prediction with CNNs. To improve the efficiency of CNNs, we introduce lean convolution operators that reduce the number of parameters and computational complexity, and can be used in a wide range of existing CNNs. Here, we exemplify their use in residual networks (ResNets), which have been very reliable for a few years now and analyzed intensively. In our experiments on three image classification problems, the proposed LeanResNet yields results that are comparable to other recently proposed reduced architectures using similar number of parameters. 
    more » « less
  3. ABSTRACT The merger of two or more galaxies can enhance the inflow of material from galactic scales into the close environments of active galactic nuclei (AGNs), obscuring and feeding the supermassive black hole (SMBH). Both recent simulations and observations of AGN in mergers have confirmed that mergers are related to strong nuclear obscuration. However, it is still unclear how AGN obscuration evolves in the last phases of the merger process. We study a sample of 60 luminous and ultra-luminous IR galaxies (U/LIRGs) from the GOALS sample observed by NuSTAR. We find that the fraction of AGNs that are Compton thick (CT; $N_{\rm H}\ge 10^{24}\rm \, cm^{-2}$) peaks at $74_{-19}^{+14}{{\ \rm per\ cent}}$ at a late merger stage, prior to coalescence, when the nuclei have projected separations (dsep) of 0.4–6 kpc. A similar peak is also observed in the median NH [$(1.6\pm 0.5)\times 10^{24}\rm \, cm^{-2}$]. The vast majority ($85^{+7}_{-9}{{\ \rm per\ cent}}$) of the AGNs in the final merger stages (dsep ≲ 10 kpc) are heavily obscured ($N_{\rm H}\ge 10^{23}\rm \, cm^{-2}$), and the median NH of the accreting SMBHs in our sample is systematically higher than that of local hard X-ray-selected AGN, regardless of the merger stage. This implies that these objects have very obscured nuclear environments, with the $N_{\rm H}\ge 10^{23}\rm \, cm^{-2}$ gas almost completely covering the AGN in late mergers. CT AGNs tend to have systematically higher absorption-corrected X-ray luminosities than less obscured sources. This could either be due to an evolutionary effect, with more obscured sources accreting more rapidly because they have more gas available in their surroundings, or to a selection bias. The latter scenario would imply that we are still missing a large fraction of heavily obscured, lower luminosity ($L_{2-10}\lesssim 10^{43}\rm \, erg\, s^{-1}$) AGNs in U/LIRGs. 
    more » « less
  4. null (Ed.)