Advanced search in
Research products
arrow_drop_down
Searching FieldsTerms
Any field
arrow_drop_down
includes
arrow_drop_down
Include:
12,302 Research products, page 1 of 1,231

  • Publications
  • Research data
  • Research software
  • 2017-2021
  • Preprint
  • IL
  • arXiv.org e-Print Archive

10
arrow_drop_down
Relevance
arrow_drop_down
  • Publication . Conference object . Article . Preprint . 2019 . Embargo End Date: 01 Jan 2019
    Open Access
    Authors: 
    Fedor Vladimirovich Borisyuk; Albert Gordo; Viswanath Sivakumar;
    Publisher: arXiv

    In this paper we present a deployed, scalable optical character recognition (OCR) system, which we call Rosetta, designed to process images uploaded daily at Facebook scale. Sharing of image content has become one of the primary ways to communicate information among internet users within social networks such as Facebook and Instagram, and the understanding of such media, including its textual information, is of paramount importance to facilitate search and recommendation applications. We present modeling techniques for efficient detection and recognition of text in images and describe Rosetta's system architecture. We perform extensive evaluation of presented technologies, explain useful practical approaches to build an OCR system at scale, and provide insightful intuitions as to why and how certain components work based on the lessons learnt during the development and deployment of the system. Comment: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD) 2018, London, United Kingdom

  • Open Access English
    Authors: 
    Gal Dalal; Balázs Szörényi; Gugan Thoppe;
    Project: NSF | RoL: FELS: RAISE: Does ev... (1840223)

    Policy evaluation in reinforcement learning is often conducted using two-timescale stochastic approximation, which results in various gradient temporal difference methods such as GTD(0), GTD2, and TDC. Here, we provide convergence rate bounds for this suite of algorithms. Algorithms such as these have two iterates, $\theta_n$ and $w_n,$ which are updated using two distinct stepsize sequences, $\alpha_n$ and $\beta_n,$ respectively. Assuming $\alpha_n = n^{-\alpha}$ and $\beta_n = n^{-\beta}$ with $1 > \alpha > \beta > 0,$ we show that, with high probability, the two iterates converge to their respective solutions $\theta^*$ and $w^*$ at rates given by $\|\theta_n - \theta^*\| = \tilde{O}( n^{-\alpha/2})$ and $\|w_n - w^*\| = \tilde{O}(n^{-\beta/2});$ here, $\tilde{O}$ hides logarithmic terms. Via comparable lower bounds, we show that these bounds are, in fact, tight. To the best of our knowledge, ours is the first finite-time analysis which achieves these rates. While it was known that the two timescale components decouple asymptotically, our results depict this phenomenon more explicitly by showing that it in fact happens from some finite time onwards. Lastly, compared to existing works, our result applies to a broader family of stepsizes, including non-square summable ones.

  • Publication . Article . Preprint . 2021 . Embargo End Date: 01 Jan 2021
    Open Access
    Authors: 
    Peizhuo Li; Kfir Aberman; Rana Hanocka; Libin Liu; Olga Sorkine-Hornung; Baoquan Chen;
    Publisher: arXiv

    Animating a newly designed character using motion capture (mocap) data is a long standing problem in computer animation. A key consideration is the skeletal structure that should correspond to the available mocap data, and the shape deformation in the joint regions, which often requires a tailored, pose-specific refinement. In this work, we develop a neural technique for articulating 3D characters using enveloping with a pre-defined skeletal structure which produces high quality pose dependent deformations. Our framework learns to rig and skin characters with the same articulation structure (e.g., bipeds or quadrupeds), and builds the desired skeleton hierarchy into the network architecture. Furthermore, we propose neural blend shapes--a set of corrective pose-dependent shapes which improve the deformation quality in the joint regions in order to address the notorious artifacts resulting from standard rigging and skinning. Our system estimates neural blend shapes for input meshes with arbitrary connectivity, as well as weighting coefficients which are conditioned on the input joint rotations. Unlike recent deep learning techniques which supervise the network with ground-truth rigging and skinning parameters, our approach does not assume that the training data has a specific underlying deformation model. Instead, during training, the network observes deformed shapes and learns to infer the corresponding rig, skin and blend shapes using indirect supervision. During inference, we demonstrate that our network generalizes to unseen characters with arbitrary mesh connectivity, including unrigged characters built by 3D artists. Conforming to standard skeletal animation models enables direct plug-and-play in standard animation software, as well as game engines. Comment: SIGGRAPH 2021. Project page: https://peizhuoli.github.io/neural-blend-shapes/ , Video: https://youtu.be/antc20EFh6k

  • Open Access English
    Authors: 
    Frédéric Paulin; Uri Shapira;
    Country: France
    Project: UKRI | Isaac Newton Institute fo... (EP/K032208/1), NSF | Mathematical Sciences Res... (1440140)

    Let $P$ be a prime polynomial in the variable $Y$ over a finite field and let $f$ be a quadratic irrational in the field of formal Laurant series in the variable $Y^{-1}$. We study the asymptotic properties of the degrees of the coefficients of the continued fraction expansion of quadratic irrationals such as $P^nf$ and prove results that are in sharp contrast to the analogue situation in zero characteristic.

  • Publication . Article . Preprint . 2019 . Embargo End Date: 01 Jan 2019
    Open Access
    Authors: 
    Itai Benjamini; Ádám Timár;
    Publisher: arXiv
    Project: EC | NOISE (772466)

    Consider an ergodic unimodular random one-ended planar graph $\G$ of finite expected degree. We prove that it has an isometry-invariant locally finite embedding in the Euclidean plane if and only if it is invariantly amenable. By "locally finite" we mean that any bounded open set intersects finitely many embedded edges. In particular, there exist invariant embeddings in the Euclidean plane for the Uniform Infinite Planar Triangulation and for the critical Augmented Galton-Watson Tree conditioned to survive. Roughly speaking, a unimodular embedding of $\G$ is one that is jointly unimodular with $\G$ when viewed as a decoration. We show that $\G$ has a unimodular embedding in the hyperbolic plane if it is invariantly nonamenable, and it has a unimodular embedding in the Euclidean plane if and only if it is invariantly amenable. Similar claims hold for representations by tilings instead of embeddings. Comment: 24 pages, 4 figures

  • Publication . Preprint . Conference object . Article . 2020
    Open Access English
    Authors: 
    Susanna F. de Rezende; Or Meir; Jakob Nordström; Toniann Pitassi; Robert Robere; Marc Vinyals;
    Project: EC | UTHOTP (279611), NSERC

    We significantly strengthen and generalize the theorem lifting Nullstellensatz degree to monotone span program size by Pitassi and Robere (2018) so that it works for any gadget with high enough rank, in particular, for useful gadgets such as equality and greater-than. We apply our generalized theorem to solve three open problems: •We present the first result that demonstrates a separation in proof power for cutting planes with unbounded versus polynomially bounded coefficients. Specifically, we exhibit CNF formulas that can be refuted in quadratic length and constant line space in cutting planes with unbounded coefficients, but for which there are no refutations in subexponential length and subpolynomial line space if coefficients are restricted to be of polynomial magnitude. •We give the first explicit separation between monotone Boolean formulas and monotone real formulas. Specifically, we give an explicit family of functions that can be computed with monotone real formulas of nearly linear size but require monotone Boolean formulas of exponential size. Previously only a non-explicit separation was known. •We give the strongest separation to-date between monotone Boolean formulas and monotone Boolean circuits. Namely, we show that the classical GEN problem, which has polynomial-size monotone Boolean circuits, requires monotone Boolean formulas of size $2^{\Omega(n/\text{polylog}(n))}$ . An important technical ingredient, which may be of independent interest, is that we show that the Nullstellensatz degree of refuting the pebbling formula over a DAG $G$ over any field coincides exactly with the reversible pebbling price of $G$ . In particular, this implies that the standard decision tree complexity and the parity decision tree complexity of the corresponding falsified clause search problem are equal. This is an extended abstract. The full version of the paper is available at https://arxiv.org/abs/2001.02144.

  • Open Access English
    Authors: 
    Yellin-Bergovoy, Ron; Heifetz, Eyal; Umurhan, Orkan M.;

    We present an explicit analysis of wave-resonant instability of swirling flows inside fast rotating cylindrical containers. The linear dynamics are decomposed into the interaction between the horizontal inner centrifugal edge waves, the outer vertical gravity waves with the aim of understanding the dynamics of the centrifugal waves. We show how the far field velocity induced respectively by the centrifugal and the gravity waves affect each other's propagation rates and amplitude growth. We follow this with an analysis of the instability in terms of a four wave interaction, two centrifugal and two gravity ones, and explain why the resonant instability can be obtained only between a pair of two counter-propagating waves, one centrifugal and one gravity. Furthermore, a near resonant regime which does not yield instability is shown to result from a phase-locking configuration between a pair of a counter-propagating centrifugal wave and a pro-propagating gravity one, where the interaction affects the waves' propagation rates but not the amplitude growth.

  • Open Access English
    Authors: 
    Ivano Baronchelli; G. Rodighiero; Harry I. Teplitz; Claudia Scarlata; Alberto Franceschini; S. Berta; Laia Barrufet; Mattia Vaccari; Matteo Bonato; Laure Ciesla; +15 more
    Publisher: HAL CCSD
    Countries: United States, France, Italy
    Project: EC | HELP (607254)

    For a sample of star forming galaxies in the redshift interval 0.15$<$z$<$0.3, we study how both the relative strength of the AGN infra-red emission, compared to that due to the star formation (SF), and the numerical fraction of AGNs, change as a function of the total stellar mass of the hosting galaxy group (M$^{*}_{\mathrm{group}}$), between $10^{10.25}$ and $10^{11.9}$M$_{\odot}$. Using a multi-component SED fitting analysis, we separate the contribution of stars, AGN torus and star formation to the total emission at different wavelengths. This technique is applied to a new multi-wavelength data-set in the SIMES field (23 not redundant photometric bands), spanning the wavelength range from the UV (GALEX) to the far-IR (Herschel) and including crucial AKARI and WISE mid-IR observations (4.5 \mu m$<\lambda<$24 \mu m), where the BH thermal emission is stronger. This new photometric catalog, that includes our best photo-z estimates, is released through the NASA/IPAC Infrared Science Archive (IRSA). Groups are identified through a friends of friends algorithm ($\sim$62% purity, $\sim$51% completeness). We identified a total of 45 galaxies requiring an AGN emission component, 35 of which in groups and 10 in the field. We find BHAR$\propto ($M$^{*}_{\mathrm{group}})^{1.21\pm0.27}$ and (BHAR/SFR)$\propto ($M$^{*}_{\mathrm{group}})^{1.04\pm0.24}$ while, in the same range of M$^{*}_{\mathrm{group}}$, we do not observe any sensible change in the numerical fraction of AGNs. Our results indicate that the nuclear activity (i.e. the BHAR and the BHAR/SFR ratio) is enhanced when galaxies are located in more massive and richer groups. Comment: 31 pages, 23 figures

  • Publication . Preprint . Article . 2018 . Embargo End Date: 01 Jan 2017
    Open Access
    Authors: 
    Amikam Levy; Anthony Kiely; J. G. Muga; Ronnie Kosloff; E. Torrontegui;
    Publisher: arXiv
    Country: Spain

    A systematic approach to design robust control protocols against the influence of different types of noise is introduced. We present control schemes which protect the decay of the populations avoiding dissipation in the adiabatic and nonadiabatic regimes and minimize the effect of dephasing. The effectiveness of the protocols is demonstrated in two different systems. Firstly, we present the case of population inversion of a two-level system in the presence of either one or two simultaneous noise sources. Secondly, we present an example of the expansion of coherent and thermal states in harmonic traps, subject to noise arising from monitoring and modulation of the control, respectively. Funding by the Israeli Science Foundation, the US Army Research Office under Contract W911NF- 15-1-0250, the Basque Government (Grant No. IT986-16), MINECO/FEDER,UE (Grants No. FIS2015-70856-P and No. FIS2015-67161-P), and QUITEMAD+CM S2013-ICE2801. 14 pags., 4 figs., 4 apps. -- Open Access funded by Creative Commons Atribution Licence 3.0

  • Publication . Article . Preprint . 2020
    Open Access
    Authors: 
    Dolev Bashi; Shay Zucker; V. Adibekyan; Nuno C. Santos; Lev Tal-Or; Trifon Trifonov; Tsevi Mazeh;
    Publisher: EDP Sciences
    Project: FCT | UIDB/04434/2020 (UIDB/04434/2020), FCT | PTDC/FIS-AST/32113/2017 (PTDC/FIS-AST/32113/2017), FCT | PTDC/FIS-AST/28953/2017 (PTDC/FIS-AST/28953/2017), FCT | UID/FIS/04434/2019 (UID/FIS/04434/2019)

    Context. The stars in the Milky Way thin and thick disks can be distinguished by several properties such as metallicity and kinematics. It is not clear whether the two populations also differ in the properties of planets orbiting the stars. In order to study this, a careful analysis of both the chemical composition and mass detection limits is required for a sufficiently large sample. Currently, this information is still limited only to large radial-velocity (RV) programs. Based on the recently published archival database of the High Accuracy Radial velocity Planet Searcher (HARPS) spectrograph, we present a first analysis of low-mass (small) planet occurrence rates in a sample of thin- and thick-disk stars. Aims. We aim to assess the effects of stellar properties on planet occurrence rates and to obtain first estimates of planet occurrence rates in the thin and thick disks of the Galaxy. As a baseline for comparison, we also aim to provide an updated value for the small close-in planet occurrence rate and compare it to results of previous RV and transit ($\textit{Kepler}$) works. Methods. We used archival HARPS RV datasets to calculate detection limits of a sample of stars that were previously analysed for their elemental abundances. For stars with known planets we first subtracted the Keplerian orbit. We then used this information to calculate planet occurrence rates according to a simplified Bayesian model in different regimes of stellar and planet properties. Results. Our results suggest that metal-poor stars and more massive stars host fewer low-mass close-in planets. We find the occurrence rates of these planets in the thin and thick disks to be comparable. In the iron-poor regimes, we find these occurrence rates to be significantly larger at the high-$\alpha$ region (thick-disk stars) as compared with the low-$\alpha$ region (thin-disk stars). In general, we find the... Comment: 10 pages, 6 figures, accepted for publication in A&A

Send a message
How can we help?
We usually respond in a few hours.