• Publicationsclose
• Other research productsclose
• 2018-2022close
• Open Access close
• Preprint close
• IL close
• arXiv.org e-Print Archive close

We quantify the luminosity contribution of active galactic nuclei (AGN) to the 12 $\mu$m, mid-infrared (MIR; 5-38 $\mu$m), and the total IR (5-1000 $\mu$m) emission in the local AGN detected in the all-sky 70-month Swift/Burst Alert Telescope (BAT) ultra hard X-ray survey. We decompose the IR spectral energy distributions (SEDs) of 587 objects into AGN and starburst components using AGN torus and star-forming galaxy templates. This enables us to recover the AGN torus emission also for low-luminosity end, down to $\log (L_{14-150}/{\rm erg}~{\rm s}^{-1}) \simeq 41$, which typically have significant host galaxy contamination. We find that the luminosity contribution of the AGN to the 12 $\mu$m, the MIR, and the total IR band is an increasing function of the 14-150 keV luminosity. We also find that for the most extreme cases, the IR pure-AGN emission from the torus can extend up to 90 $\mu$m. The obtained total IR AGN luminosity through the IR SED decomposition enables us to estimate the fraction of the sky obscured by dust, i.e., the dust covering factor. We demonstrate that the median of the dust covering factor is always smaller than that of the X-ray obscuration fraction above the AGN bolometric luminosity of $\log (L_{\rm bol}/{\rm erg}~{\rm s}^{-1}) \simeq 42.5$. Considering that X-ray obscuration fraction is equivalent to the covering factor coming from both the dust and gas, it indicates that an additional neutral gas component, along with the dusty torus, is responsible for the absorption of X-ray emission. Comment: 21 pages, 15 figures, accepted for publication in ApJ. The full list of Table 1 is available at http://www.kusastro.kyoto-u.ac.jp/~ichikawa/Table1_MR_20181107.txt

Deep Neural Networks have recently gained lots of success after enabling several breakthroughs in notoriously challenging problems. Training these networks is computationally expensive and requires vast amounts of training data. Selling such pre-trained models can, therefore, be a lucrative business model. Unfortunately, once the models are sold they can be easily copied and redistributed. To avoid this, a tracking mechanism to identify models as the intellectual property of a particular vendor is necessary. In this work, we present an approach for watermarking Deep Neural Networks in a black-box way. Our scheme works for general classification tasks and can easily be combined with current learning algorithms. We show experimentally that such a watermark has no noticeable impact on the primary task that the model is designed for and evaluate the robustness of our proposal against a multitude of practical attacks. Moreover, we provide a theoretical analysis, relating our approach to previous work on backdooring.

We construct a universally Bayes consistent learning rule that satisfies differential privacy (DP). We first handle the setting of binary classification and then extend our rule to the more general setting of density estimation (with respect to the total variation metric). The existence of a universally consistent DP learner reveals a stark difference with the distribution-free PAC model. Indeed, in the latter DP learning is extremely limited: even one-dimensional linear classifiers are not privately learnable in this stringent model. Our result thus demonstrates that by allowing the learning rate to depend on the target distribution, one can circumvent the above-mentioned impossibility result and in fact, learn \emph{arbitrary} distributions by a single DP algorithm. As an application, we prove that any VC class can be privately learned in a semi-supervised setting with a near-optimal \emph{labeled} sample complexity of $\tilde{O}(d/\varepsilon)$ labeled examples (and with an unlabeled sample complexity that can depend on the target distribution).

Optical diffraction limit has been a long-term scientific issue since Ernst Abbe first introduced the concept in 1873. It is a constraint on the smallest light spot that can be achieved. Substantial effort has been invested in the past decade to beat this limit by exploiting evanescent waves. But this method encounters serious near-field limitations. A more promising route to breaking the constraint is to explore optical superoscillation in the far field with engineered metamaterials. However, these particular structures involve with very complicated optimization-based design that requires precisely tailoring the interference of propagating waves with low spatial frequency. To overcome these limitations, here we explore a new approach based on the two-hundred-year-old discovery: Possion-Arago spots. We show for the first time that by using a single disc, constructive interference of propagating waves with high-spatial-frequency wavevectors can be realized, generating a diffraction-unlimited localized Possion-Arago spot with achievable size down to $\lambda/$20. Actually, such an element permits creation of an ultra-long nearly nondiffracting superoscillatory needle with appreciable field of view. This easy-to-fabrication element provides a promising route to overcome the diffraction limit, thus might open new avenues to exploit various applications in different fields. Comment: The paper includes 4 figures

We present a joint experimental and theoretical study of spin dynamics of a single $^{88}$Sr$^+$ ion colliding with an ultracold cloud of Rb atoms in various hyperfine states. While spin-exchange between the two species occurs after 9.1(6) Langevin collisions on average, spin-relaxation of the Sr$^+$ ion Zeeman qubit occurs after 48(7) Langevin collisions which is significantly slower than in previously studied systems due to a small second-order spin-orbit coupling. Furthermore, a reduction of the endothermic spin-exchange rate was observed as the magnetic field was increased. Interestingly, we found that, while the phases acquired when colliding on the spin singlet and triplet potentials vary largely between different partial waves, the singlet-triplet phase difference, which determines the spin-exchange cross-section, remains locked to a single value over a wide range of partial-waves which leads to quantum interference effects. Comment: 5 pages, 5 figures and Supplemental Material

Random Walk (RW) is a common numerical tool for modeling the Advection-Diffusion equation. In this work, we develop a second order scheme for incorporating a heterogeneous reaction (i.e., a Robin boundary condition) in the RW model. In addition, we apply the approach in two test cases. We compare the second order scheme with the first order one as well as with analytical and other numerical solution. We show that the new scheme can reduce the computational error significantly, relative to the first order scheme. This reduction comes at no additional computational cost.

Convolutional Neural Networks (CNNs) have become common in many fields including computer vision, speech recognition, and natural language processing. Although CNN hardware accelerators are already included as part of many SoC architectures, the task of achieving high accuracy on resource-restricted devices is still considered challenging, mainly due to the vast number of design parameters that need to be balanced to achieve an efficient solution. Quantization techniques, when applied to the network parameters, lead to a reduction of power and area and may also change the ratio between communication and computation. As a result, some algorithmic solutions may suffer from lack of memory bandwidth or computational resources and fail to achieve the expected performance due to hardware constraints. Thus, the system designer and the micro-architect need to understand at early development stages the impact of their high-level decisions (e.g., the architecture of the CNN and the amount of bits used to represent its parameters) on the final product (e.g., the expected power saving, area, and accuracy). Unfortunately, existing tools fall short of supporting such decisions. This paper introduces a hardware-aware complexity metric that aims to assist the system designer of the neural network architectures, through the entire project lifetime (especially at its early stages) by predicting the impact of architectural and micro-architectural decisions on the final product. We demonstrate how the proposed metric can help evaluate different design alternatives of neural network models on resource-restricted devices such as real-time embedded systems, and to avoid making design mistakes at early stages.

Two dimensional channel flow simulations of FENE-P fluid in the elasto-inertial turbulence regime reveal distinct regimes ranging from chaos to a steady travelling wave which takes the form of an arrowhead structure. This coherent structure provides new insights in the polymer/flow interactions driving EIT, which are observed in a set of controlled numerical experiments and the study of transfer between elastic and turbulent kinetic energy. Comment: Revised version under consideration for PRF

The current leading computer vision models are typically feed forward neural models, in which the output of one computational block is passed to the next one sequentially. This is in sharp contrast to the organization of the primate visual cortex, in which feedback and lateral connections are abundant. In this work, we propose a computational model for the role of lateral connections in a given block, in which the weights of the block vary dynamically as a function of its activations, and the input from the upstream blocks is iteratively reintroduced. We demonstrate how this novel architectural modification can lead to sizable gains in performance, when applied to visual action recognition without pretraining and that it outperforms the literature architectures with recurrent feedback processing on ImageNet.

Let $\lambda$ denote the Liouville function. We show that, as $X \rightarrow \infty$, $$\int_{X}^{2X} \sup_{\substack{P(Y)\in \mathbb{R}[Y]\\ deg(P)\leq k}} \Big | \sum_{x \leq n \leq x + H} \lambda(n) e(-P(n)) \Big |\ dx = o ( X H)$$ for all fixed $k$ and $X^{\theta} \leq H \leq X$ with $0 < \theta < 1$ fixed but arbitrarily small. Previously this was only established for $k \leq 1$. We obtain this result as a special case of the corresponding statement for (non-pretentious) $1$-bounded multiplicative functions that we prove. In fact, we are able to replace the polynomial phases $e(-P(n))$ by degree $k$ nilsequences $\overline{F}(g(n) \Gamma)$. By the inverse theory for the Gowers norms this implies the higher order asymptotic uniformity result $$\int_{X}^{2X} \| \lambda \|_{U^{k+1}([x,x+H])}\ dx = o ( X )$$ in the same range of $H$. We present applications of this result to patterns of various types in the Liouville sequence. Firstly, we show that the number of sign patterns of the Liouville function is superpolynomial, making progress on a conjecture of Sarnak about the Liouville sequence having positive entropy. Secondly, we obtain cancellation in averages of $\lambda$ over short polynomial progressions $(n+P_1(m),\ldots, n+P_k(m))$, which in the case of linear polynomials yields a new averaged version of Chowla's conjecture. We are in fact able to prove our results on polynomial phases in the wider range $H\geq \exp((\log X)^{5/8+\varepsilon})$, thus strengthening also previous work on the Fourier uniformity of the Liouville function. Comment: 107 pages; to appear in Ann. of Math